WEBVTT 00:00:00.020 --> 00:00:04.120 If you've ever been to PyCon, you know one of the best parts of the expo hall is Startup Row, 00:00:04.270 --> 00:00:07.880 a stretch of booths where early-stage companies built on Python show off what they're creating, 00:00:08.360 --> 00:00:12.980 but only attendees get to walk that lane. So let's bring it to everyone. In this episode, 00:00:13.190 --> 00:00:17.760 we stroll down Startup Lane together. We kick things off with the organizers, Jason and Shay, 00:00:18.240 --> 00:00:24.019 who share the program's origin story going back to Paul Graham and the PSF, plus some surprising 00:00:24.040 --> 00:00:30.760 stats, including two unicorns among the alumni. Then we meet five startups, Tetrix, bringing AI 00:00:31.000 --> 00:00:36.220 to institutional investing in private markets, ArcGET, security that lives inside your app as an 00:00:36.310 --> 00:00:42.660 SDK, Femoral.dev, serverless hosting built for Python web apps, Capitio, an identity and authority 00:00:42.840 --> 00:00:48.820 layer for AI agents, and Pixel Table, a multimodal database from Marcel Kroniker, co-creator of Apache 00:00:49.020 --> 00:00:54.000 Parquet. See if you can spot the theme running through them all. Let's go for a walk. This is 00:00:54.020 --> 00:00:58.160 Episode 551, recorded June 11th, 2026. 00:01:15.600 --> 00:01:20.160 Welcome to Talk Python To Me, the number one Python podcast for developers and data scientists. 00:01:20.760 --> 00:01:22.080 This is your host, Michael Kennedy. 00:01:22.140 --> 00:01:26.000 I'm a PSF fellow who's been coding for over 25 years. 00:01:26.640 --> 00:01:27.760 Let's connect on social media. 00:01:28.080 --> 00:01:31.240 You'll find me and Talk Python on Mastodon, BlueSky, and X. 00:01:31.400 --> 00:01:33.340 The social links are all in your show notes. 00:01:34.120 --> 00:01:37.640 You can find over 10 years of past episodes at talkpython.fm. 00:01:37.820 --> 00:01:41.000 And if you want to be part of the show, you can join our recording live streams. 00:01:41.280 --> 00:01:41.840 That's right. 00:01:42.080 --> 00:01:45.300 We live stream the raw uncut version of each episode on YouTube. 00:01:45.880 --> 00:01:50.300 Just visit talkpython.fm/youtube to see the schedule of upcoming events. 00:01:50.480 --> 00:01:54.180 Be sure to subscribe there and press the bell so you'll get notified anytime we're recording. 00:01:54.860 --> 00:01:57.840 What if your AI agents worked like FastAPI microservices, 00:01:58.500 --> 00:02:00.780 typed, autonomous, and discovering each other at runtime? 00:02:01.480 --> 00:02:03.140 That's the world Agent Field is building. 00:02:03.810 --> 00:02:06.660 Join them at talkpython.fm/agentfield. 00:02:07.860 --> 00:02:13.500 To kick off this episode, we have the organizers of Startup Bro at PyCon, Jason and Shay. 00:02:14.360 --> 00:02:15.300 Welcome to Talk Python To Me. 00:02:15.610 --> 00:02:16.080 Great to meet you. 00:02:16.320 --> 00:02:17.100 Thanks for having us. 00:02:17.540 --> 00:02:18.080 Great to be here. 00:02:18.340 --> 00:02:47.240 I think this is a really cool idea. Long ago, way back, I think it was 2023, I did a stroll down startup lane. And the idea was, hey, this startup row is a really cool experience. But only attendees of the conference actually get to walk down the lane, walk down startup row, talk to each company, see what they're building, get that energy. So let's bring it to the podcast listeners, bring it to the world. So here we are. And you two are the organizers, right? 00:02:47.340 --> 00:02:48.500 Mm-hmm. That's right. 00:02:48.790 --> 00:02:50.560 Well, thank you very much on behalf of everyone. 00:02:50.930 --> 00:02:51.520 Quick introduction. 00:02:52.480 --> 00:02:53.340 Shay, welcome to the show. 00:02:53.690 --> 00:02:54.080 Thank you. 00:02:54.520 --> 00:02:55.140 Yeah, tell us about yourself. 00:02:55.480 --> 00:02:57.780 So my background's in early stage venture and technology. 00:02:57.990 --> 00:03:00.460 I was part of the founding team of a fund called True Ventures. 00:03:00.920 --> 00:03:03.160 And then I was there for three funds, seven years. 00:03:03.820 --> 00:03:06.660 And then after that, decided I wanted to start a company of my own. 00:03:07.040 --> 00:03:12.120 So we were a full stack Python dev shop, and we were building an EdTech SaaS product. 00:03:12.560 --> 00:03:18.060 And we had the opportunity to be featured on Startup Row at PyCon in Montreal in 2015. 00:03:18.340 --> 00:03:20.160 So that was my introduction to the community. 00:03:20.540 --> 00:03:21.800 And I had a fantastic experience. 00:03:22.480 --> 00:03:23.700 And then we later got acquired. 00:03:24.280 --> 00:03:29.040 And so in order to give back to the community, I started judging with Jason a couple years 00:03:29.240 --> 00:03:29.320 later. 00:03:29.380 --> 00:03:33.240 He asked me if I would help contribute to the process by being one of the judges that 00:03:33.400 --> 00:03:34.020 selected the companies. 00:03:34.680 --> 00:03:37.260 And then a couple of years ago, asked me if I would co-organize with him. 00:03:37.700 --> 00:03:39.020 So that's how I got involved. 00:03:39.220 --> 00:03:41.800 I've loved being part of the community. 00:03:41.920 --> 00:03:47.200 I think it's fantastic and used to go to software development meetups in San Francisco to recruit technical talent. 00:03:47.660 --> 00:03:51.360 And I always wondered, you know, with my venture investing hat on, why aren't the VCs here? 00:03:51.480 --> 00:03:53.620 Because these are the people building incredible technologies. 00:03:53.920 --> 00:04:00.020 And so really see the opportunity to feature these great companies that have become part of our corpus of startup for talent. 00:04:00.560 --> 00:04:03.760 So you've lived the entire journey, which is awesome. Congratulations. 00:04:04.740 --> 00:04:05.080 Thank you. 00:04:05.580 --> 00:04:06.260 Jason, welcome. 00:04:06.620 --> 00:04:07.780 Hey, good to be here. 00:04:08.740 --> 00:04:09.980 So how did I get involved? 00:04:09.980 --> 00:04:21.579 It was around 2015 when I was a financial journalist primarily covering startups and technology and venture capital stuff. 00:04:21.840 --> 00:04:35.820 When at the time, the then organizer of Startup Road asked me to pinch hit for a friend of mine who was going to be co-hosting or emceeing a pitch event in Chicago. 00:04:38.020 --> 00:04:40.900 And unfortunately, my friend wasn't able to participate. 00:04:43.340 --> 00:04:49.460 And at the time, the organizer, Startup Pro, asked me, oh, hey, Jason, you do some stuff with media. 00:04:49.700 --> 00:04:51.980 You seem like you'd be good on stage. 00:04:52.100 --> 00:04:53.560 I'm like, no, that's not true. 00:04:54.380 --> 00:04:56.620 And he's just like, but you know some stuff about startups. 00:04:56.660 --> 00:04:57.820 And I'm like, yeah, that is true. 00:04:57.840 --> 00:05:07.300 And long story short is I got convinced to host this pitch event, and I really fell in love with the Python community. 00:05:07.360 --> 00:05:12.780 This was during a time in my career where I was sort of transitioning into doing more like data journalism. 00:05:13.640 --> 00:05:20.520 And I got really frustrated with like the limits of what I could do with, you know, Excel, you know. 00:05:20.720 --> 00:05:27.040 And this was just at the very beginning of my journey, learning what little Python that I do know. 00:05:27.980 --> 00:05:30.680 And so, you know, it's an amazing community. 00:05:30.740 --> 00:05:37.160 and very long story short, this is my 10th year of organizing or being a co-organizer of Startup 00:05:37.190 --> 00:05:43.660 Row. I sort of became the primary organizer in 2019, and I'm really excited to have Shea board. 00:05:45.040 --> 00:05:48.060 I have some statistics to share, but I'm happy to share those later. 00:05:50.280 --> 00:05:52.880 But yeah, that's a little bit about my background. Awesome. Very interesting. 00:05:53.320 --> 00:05:58.040 So let's, I do want to hear your statistics. I think they're very interesting. And I have a 00:05:58.000 --> 00:06:03.420 little hint of what we discussed. So I have a hint of where you're going. But before we get into 00:06:03.440 --> 00:06:08.880 that, let's just give everyone a quick, Shay, maybe give everyone a quick overview. What is Startup Row? 00:06:09.420 --> 00:06:13.960 Why are companies there? How do they get there? Yeah, it's a great backstory, actually. So in 2011, 00:06:14.380 --> 00:06:19.340 Paul Graham of Y Combinator was collaborating with the Python Software Foundation. And with the 00:06:19.520 --> 00:06:24.679 impetus of we want to see more startups at PyCon, we want to feature more of these amazing companies 00:06:24.680 --> 00:06:27.040 that are starting new technologies and building things from scratch. 00:06:27.620 --> 00:06:29.580 We want to give them an opportunity for a spotlight, 00:06:29.760 --> 00:06:32.560 but many of them are so early stage they can't afford the price of admission 00:06:32.740 --> 00:06:33.520 or the conference ticket. 00:06:33.760 --> 00:06:37.560 What can we do to help create an on-ramp for them and then to feature them? 00:06:37.980 --> 00:06:39.340 And that's the origin of how it started. 00:06:39.660 --> 00:06:45.080 And it's been amazing that we're 15 years in and the company is just fantastic. 00:06:46.420 --> 00:06:50.160 As Jason alluded, we've got some great alumni companies as well. 00:06:51.360 --> 00:06:53.620 All right. Well, let's hear the stats, Jason. 00:06:54.200 --> 00:07:02.360 Okay, so since the program's inception, there's been approximately 170, 175 companies that have come through Startup Pro. 00:07:03.819 --> 00:07:06.860 The format of the program has changed a little bit over time. 00:07:07.240 --> 00:07:19.400 It used to be the case that there would be eight companies featured, say, on Friday, and then a whole brand new batch of eight companies featured on Saturday, the two different days of the primary expo hall. 00:07:19.980 --> 00:07:25.180 That was kind of a crazy and untenable situation for everybody in the long haul. 00:07:25.520 --> 00:07:27.960 And so we have now eight companies per batch. 00:07:28.100 --> 00:07:35.660 And so if you look at from 2019 through 2025, so that's counted six batches. 00:07:36.860 --> 00:07:39.620 You know, we're looking at about eight companies. 00:07:39.680 --> 00:07:40.460 Oh, hang on one second. 00:07:40.860 --> 00:07:41.700 Was that six batches? 00:07:42.180 --> 00:07:42.420 One moment. 00:07:42.420 --> 00:07:42.940 I apologize. 00:07:43.460 --> 00:07:44.020 Seven batches. 00:07:44.140 --> 00:07:44.540 Forgive me. 00:07:45.800 --> 00:07:46.560 Seven batches. 00:07:46.700 --> 00:07:50.700 We have roughly 60 companies in that pool. 00:07:51.910 --> 00:07:55.600 Of those, we have 32 of them are currently active. 00:07:56.600 --> 00:07:58.100 11 of them have been acquired. 00:07:59.000 --> 00:08:01.680 15 of them are unfortunately no longer with us. 00:08:02.520 --> 00:08:07.380 And a couple of them, based on public activity, a little bit difficult to tell where they're at. 00:08:08.040 --> 00:08:14.580 But the bottom line is that of this population of companies that have come through recent startup row batches, 00:08:15.500 --> 00:08:22.460 You know, we're looking at approximately, you know, 12 to, you know, 15 percent of them having been acquired. 00:08:23.200 --> 00:08:30.140 And then also in that group, we have a couple of companies that have achieved two companies that have achieved unicorn status. 00:08:30.270 --> 00:08:32.680 So evaluation of a billion dollars or more. 00:08:33.440 --> 00:08:46.440 That includes a returning PyCon sponsor, ChainGuard, which is, I think, valued at three and a half or three point eight billion dollars today after their most recent funding round. 00:08:47.620 --> 00:08:54.660 And, yeah, it's it's a it's a pretty impressive population of companies that have come through Startup Row. 00:08:55.500 --> 00:09:02.720 And it's been a real joy to see, you know, where everybody has gone after their, you know, after their time at PyCon. 00:09:03.080 --> 00:09:15.500 That's incredible. I think those are super good numbers compared to just the general population of startups. Maybe we could kind of close it out with your thoughts, Shay, since you have some of the investing side. Those sound like good numbers to you as well? 00:09:15.800 --> 00:09:29.640 That's fantastic. And one of the things I enjoy is that it's such a robust community that when a company comes into Startup Bro, they're getting to meet with their ICP, their ideal customer profile. They're getting to hear straight from the mouths of the developers. 00:09:30.620 --> 00:09:32.100 They're also fantastic people. 00:09:32.200 --> 00:09:41.280 So it's really exciting to help connect them with the right opportunities and to see some of the larger sponsors at PyCon come over and interface with them and learn about what they're developing. 00:09:41.460 --> 00:09:48.300 And one of my favorite times at Startup Pro is to see everybody gathered around the computer screen watching the demo and pointing and troubleshooting. 00:09:49.380 --> 00:09:56.420 And one of the companies that you've interviewed, they were saying how many uploads they've had of their product even just during the two days that they were featured. 00:09:56.620 --> 00:09:59.440 And so you see the excitement straight from the developers. 00:10:00.200 --> 00:10:07.720 It doesn't hurt that it's become the lingua franca for all of the large language models and the different evolution of the technology that we're seeing today. 00:10:08.020 --> 00:10:10.140 So Jason, I think it's a fantastic opportunity. 00:10:10.580 --> 00:10:14.440 And we love being part of helping the founders on their journey. 00:10:14.820 --> 00:10:17.140 Well, let's see if we can get some extra work for you all next year. 00:10:17.260 --> 00:10:21.780 Maybe you all could give a quick shout out just to how people apply and what the criteria are. 00:10:21.980 --> 00:10:23.360 Oh, absolutely. 00:10:24.220 --> 00:10:24.880 Everyone's take it. 00:10:25.380 --> 00:10:25.900 Oh, sorry. 00:10:26.060 --> 00:10:26.739 Oh, sure. 00:10:28.699 --> 00:10:29.100 Sure. 00:10:29.340 --> 00:10:29.500 Absolutely. 00:10:29.980 --> 00:10:35.340 So there are three very basic rules for qualifying to be on Startup Row. 00:10:36.400 --> 00:10:38.820 And two of them are somewhat fuzzy. 00:10:39.350 --> 00:10:43.440 The one non-negotiable is obviously you got to use Python somewhere in your stack. 00:10:43.890 --> 00:10:44.560 The more the better. 00:10:44.870 --> 00:10:51.800 There is definitely a slight implicit bias in favor of companies that are building open source projects. 00:10:52.030 --> 00:10:54.980 But that is by no means a requirement to be on Startup Row. 00:10:55.500 --> 00:11:00.480 In addition to that, in general, companies are younger than sort of two and a half or 00:11:00.590 --> 00:11:01.400 three years old. 00:11:01.790 --> 00:11:07.020 And that can be measured from the, you know, either from the incorporation date or like 00:11:07.140 --> 00:11:11.440 if the team has been working on a project and then did some major pivot like a couple 00:11:11.580 --> 00:11:16.160 years ago, you know, that's not necessarily a count against their, you know, their eligibility. 00:11:17.040 --> 00:11:20.999 It's just like, you know, we're trying not to feature companies that were founded 10 00:11:21.020 --> 00:11:26.020 years ago and are, you know, more justifiably categorized as like robust small software 00:11:26.260 --> 00:11:26.440 businesses. 00:11:27.000 --> 00:11:27.120 Right. 00:11:27.640 --> 00:11:34.860 And then the final one is, you know, in general, roughly 25 people on the team or smaller, 00:11:35.440 --> 00:11:39.720 again, with a bias in favor of smaller teams, just in service of there are plenty. 00:11:40.000 --> 00:11:45.400 I'm sure that there are plenty, you know, 120 person Series B startups that would love 00:11:45.560 --> 00:11:48.500 free booth space on startup pro at PyCon US. 00:11:48.560 --> 00:11:52.460 But this really is for much smaller, earlier stage, high growth ventures. 00:11:52.760 --> 00:11:53.320 That sounds great. 00:11:53.690 --> 00:11:56.120 Final word, Shay, how do people apply? 00:11:56.360 --> 00:11:57.860 When do applications open? 00:11:58.040 --> 00:11:58.600 Things like that. 00:11:58.840 --> 00:12:03.420 Yes, it moves a little bit depending upon our collaboration with the Python Software Foundation. 00:12:03.820 --> 00:12:09.220 But usually applications open right around December, beginning of the year, because PyCon, 00:12:09.290 --> 00:12:10.260 as you know, is held in May. 00:12:10.330 --> 00:12:12.940 And so we like to have plenty of lead time for the companies to apply. 00:12:13.210 --> 00:12:15.940 And also we're trying to coordinate better with the speaker talks. 00:12:16.200 --> 00:12:20.940 They can do a submission to have a speaker talk as well, but usually around December, early December. 00:12:21.520 --> 00:12:28.260 And Jason, we should probably put an email address up so people can start to reach out in advance if they'd like as well. 00:12:28.560 --> 00:12:36.020 Yes, we should. I do believe we have startups at python.org or it's startupro at python.org. 00:12:36.220 --> 00:12:38.580 But Michael, I'll get back to you and confirm. 00:12:39.020 --> 00:12:43.740 Let's do this. Send it to me and I'll put it in the show notes so people have it right there. 00:12:44.020 --> 00:12:44.560 That would be great. 00:12:45.100 --> 00:12:50.480 Once we put the call out, it's a quick intake application, a few short questions, and then 00:12:50.640 --> 00:12:53.240 we have a panel of judges with different backgrounds to evaluate. 00:12:53.880 --> 00:12:57.600 And the news usually goes out in January and then tee everything up from there. 00:12:57.760 --> 00:12:58.120 Fantastic. 00:12:58.600 --> 00:12:58.920 All right. 00:12:59.200 --> 00:13:00.700 Shay, Jason, thank you for being on the show. 00:13:01.000 --> 00:13:02.660 And thanks for organizing Startup Row. 00:13:02.760 --> 00:13:03.100 It's been great. 00:13:03.420 --> 00:13:04.040 Thank you, Michael. 00:13:04.180 --> 00:13:04.660 Great to meet you. 00:13:05.080 --> 00:13:05.220 Yeah. 00:13:05.520 --> 00:13:05.860 Yeah, you bet. 00:13:05.960 --> 00:13:06.140 Bye-bye. 00:13:06.480 --> 00:13:06.680 Cheers. 00:13:08.740 --> 00:13:11.320 This portion of Talk Python on Me is brought to you by Agent Field. 00:13:11.660 --> 00:13:16.760 What happens when you give hundreds of AI agents a shared code base and let them write code, 00:13:17.190 --> 00:13:19.440 review each other's work, and ship to production? 00:13:20.220 --> 00:13:23.680 Well, that's exactly what the team behind AgentField AI built. 00:13:24.300 --> 00:13:28.040 And the wild part, it's not some proprietary system locked behind a paywall. 00:13:28.380 --> 00:13:30.100 It's an open source Python library. 00:13:30.600 --> 00:13:34.340 Now, where most agent frameworks have you wiring up DAGs and workflows, 00:13:35.040 --> 00:13:39.600 AgentField lets you build AI agents the way you'd build FastAPI microservices. 00:13:40.260 --> 00:13:43.740 Think typed Python functions that become autonomous services. 00:13:44.540 --> 00:13:49.620 They discover each other at runtime, call each other like APIs, scale independently, 00:13:50.180 --> 00:13:52.060 fail independently, and recover on their own. 00:13:52.320 --> 00:13:53.040 And here's the thing. 00:13:53.440 --> 00:13:55.460 You're not just orchestrating LLM calls. 00:13:56.080 --> 00:13:58.200 You can orchestrate entire anonymous tools. 00:13:58.840 --> 00:14:04.280 Spin up multiple Claude Code instances, codec sessions, any coding harness you want, 00:14:04.560 --> 00:14:09.620 all running as live nodes on the same architecture, collaborating and verifying each other's output. 00:14:10.440 --> 00:14:14.720 that's how they build the factory and it's completely free and open source check it out 00:14:14.730 --> 00:14:21.080 at talkpython.fm/agentfield that's talkpython.fm/agentfield the link is in your podcast 00:14:21.210 --> 00:14:26.840 player show notes thank you to agentfield for supporting the show now let's meet our first 00:14:27.270 --> 00:14:34.240 startup tetrix from nanid and grant nanid grant welcome to talk python and me amazing to have you 00:14:34.320 --> 00:14:38.779 here thank you for having us michael yeah congratulations on being part of startup row 00:14:39.060 --> 00:14:44.760 A few years ago, I did this episode where we did a stroll down startup lane, and I'm very excited to do it again this year. 00:14:45.260 --> 00:14:50.320 And there's a theme. We'll see if people can detect the theme throughout this episode. 00:14:50.780 --> 00:14:53.640 But it is also evident here, and that's great. 00:14:53.880 --> 00:14:59.600 So let's just do a quick round of introductions before we dive into Tetrix, your company, and how you got on Startup Row and stuff. 00:14:59.880 --> 00:15:00.360 Nanit, go first. 00:15:00.860 --> 00:15:05.380 Sure, happy to. So I'm Nanit. I'm the co-founder and chief technology officer of Tetrix. 00:15:05.620 --> 00:15:11.920 I've been a huge nerd about tech ever since I was a child and transformed that into my degree, then eventually a career. 00:15:12.360 --> 00:15:16.220 And now I get to work with some of the smartest people building some of the coolest things in financial markets. 00:15:16.780 --> 00:15:21.060 Very cool. It's such an incredible time to be a builder. So amazing. Grant. Hello. 00:15:21.440 --> 00:15:25.240 Yeah. Hi. So I'm Grant. I'm the founding engineer at Tetrix. 00:15:26.460 --> 00:15:32.060 Background in finance and technology. Former investment banker turned software engineer. 00:15:32.780 --> 00:15:38.820 So really excited to, was really excited to join Tetrix and build what we have. 00:15:39.020 --> 00:15:44.380 Well, I have a really good friend who has worked in basically as a lawyer and in banking and so on. 00:15:44.680 --> 00:15:45.740 And he also does tech now. 00:15:45.800 --> 00:15:47.360 We love to exchange stories. 00:15:47.540 --> 00:15:48.580 It's a different kind of world, huh? 00:15:48.980 --> 00:15:49.260 Very. 00:15:50.940 --> 00:15:51.980 I think it's a fun one though. 00:15:52.100 --> 00:15:55.400 And, you know, let's talk about your project that you've been working on, Tetrix. 00:15:55.920 --> 00:15:57.940 So this is what you were presenting on Startup Row. 00:15:58.280 --> 00:15:59.540 I'll tell it to the audience. 00:15:59.620 --> 00:16:00.020 What do you got? 00:16:00.300 --> 00:16:00.620 Absolutely. 00:16:01.060 --> 00:16:05.460 So Tetrix is an AI investment platform for institutional investors investing in private 00:16:05.660 --> 00:16:05.940 markets. 00:16:06.440 --> 00:16:09.220 And I know that's a mouthful for people who aren't in this space. 00:16:09.380 --> 00:16:10.760 So let me break it down a little bit. 00:16:11.420 --> 00:16:13.800 So AI investment platform for institutional investors. 00:16:14.220 --> 00:16:19.460 So our customers tend to be institutional investors like endowments, foundations, family 00:16:19.620 --> 00:16:20.740 offices, pension funds. 00:16:21.110 --> 00:16:24.580 Think of it as the pool of investors with the largest amounts of capital in the world. 00:16:24.960 --> 00:16:28.680 And when you have that much capital, you're not just invested in public markets, which 00:16:28.600 --> 00:16:33.140 is stuff like stocks, bonds, things that people are generally familiar with. But you also invest 00:16:33.140 --> 00:16:37.820 in private markets, which is venture capital, private equity, private credit, private infrastructure, 00:16:37.820 --> 00:16:43.660 and those asset classes. And what Tetrix does is it makes the management of your investments in 00:16:43.720 --> 00:16:48.820 private markets much easier by solving three main pain points. Pain point one is the pain of 00:16:49.060 --> 00:16:53.240 monitoring and collecting documents from multiple different sources on the internet. Pain point 00:16:53.380 --> 00:16:58.040 number two is taking all of that unstructured data, structuring it and normalizing it, 00:16:58.300 --> 00:17:02.680 And then solving pain point three, which is the lack of analytics and insights that currently 00:17:02.860 --> 00:17:07.280 exist in the market by bridging that with our data sets and our extractions to make sure 00:17:07.400 --> 00:17:10.560 that institutional investors can make more confident investment decisions. 00:17:11.660 --> 00:17:14.079 And the platform has really taken off since we launched it. 00:17:14.120 --> 00:17:18.280 And the ROI is very evident, whether it's compression of time from 45 days to one day 00:17:18.400 --> 00:17:23.040 to insights, whether it is the thousands of hours manually being saved or the 10x deeper 00:17:23.319 --> 00:17:25.140 insights that customers are generating from the platform. 00:17:25.520 --> 00:17:25.699 Okay. 00:17:26.040 --> 00:17:27.220 Well, that sounds super interesting. 00:17:27.420 --> 00:17:32.940 I did need to, when I first spoke to you, get a little bit of education on what limited partners exactly means in investing. 00:17:33.340 --> 00:17:37.980 If I think about where AI has leverage, people talk about, oh, the AI bubble. 00:17:38.740 --> 00:17:42.860 Maybe like my male client doesn't necessarily need AI and that kind of just gets in the way. 00:17:43.300 --> 00:17:45.980 But there are areas where it applies so well. 00:17:46.290 --> 00:17:48.980 I think coding and data science is probably number one. 00:17:49.320 --> 00:17:55.860 But I've always felt that investing has this special lever for AI that you could really get a lot out of. 00:17:56.040 --> 00:17:57.740 And I haven't done nothing with it. 00:17:57.760 --> 00:17:58.320 I don't know why. 00:17:58.360 --> 00:18:00.060 I feel like, I guess because I don't have the experience, 00:18:00.280 --> 00:18:02.020 but Grant, you've been in a banker. 00:18:02.520 --> 00:18:05.160 So you know the actual day-to-day problems, right? 00:18:05.500 --> 00:18:06.060 Yeah, exactly. 00:18:06.360 --> 00:18:08.300 I mean, one of the, like, I mean, as a banker, 00:18:08.620 --> 00:18:11.140 like doing what we do, like docking processing, 00:18:11.500 --> 00:18:15.860 like that was all manual effort of like literally typing in like values 00:18:16.000 --> 00:18:18.800 into Excel from like PDF reports, those types of things. 00:18:19.460 --> 00:18:22.940 And being able to kind of like streamline that really manual process 00:18:23.400 --> 00:18:25.280 is like really fascinating. 00:18:25.440 --> 00:18:30.320 Because there were so many hours spent late at night doing this very annoying manual process. 00:18:30.610 --> 00:18:36.700 And now that you can kind of do that unstructured to structured transition with LLMs is really exciting. 00:18:36.940 --> 00:18:37.680 It's super exciting. 00:18:38.120 --> 00:18:44.820 And people always say, oh, man, I always wanted to have some kind of thing help me out. 00:18:45.010 --> 00:18:46.500 But I wanted to help me with the drudgery. 00:18:46.550 --> 00:18:48.540 I didn't want it to write music and write code. 00:18:48.590 --> 00:18:51.260 I wanted to do the boring stuff and I wanted to do the exciting stuff. 00:18:51.490 --> 00:18:54.180 But it sounds like you're kind of leveraging it for the boring stuff here. 00:18:54.560 --> 00:18:54.960 Exactly. 00:18:55.360 --> 00:19:00.460 Exactly. And even not just for customer specific use cases, when you look and peer behind the 00:19:00.590 --> 00:19:04.520 curtain at Tetrix internally, whether we're on the engineering side, the business side, 00:19:04.960 --> 00:19:08.600 everyone is leveraging what's being shipped out, whether it's connecting to MCP servers to make 00:19:08.600 --> 00:19:13.260 their workflows more efficient, whether it is using cloud routines to automate some of the tasks. 00:19:13.530 --> 00:19:17.520 We are always thinking about how can we execute faster internally and how can we keep delivering 00:19:17.700 --> 00:19:22.960 value to our customers using the AI as a tool rather than just kind of AI wash everything and 00:19:22.960 --> 00:19:24.960 say we are an AI native company just for the sake of it. 00:19:25.300 --> 00:19:27.480 I mean, that's where the VC is, but just kidding. 00:19:28.100 --> 00:19:29.220 Sort of, sort of. 00:19:29.740 --> 00:19:33.680 But how do you guys go about making sure that the AI stays on track? 00:19:34.400 --> 00:19:39.920 You know, one, I actually, I did use AI for this investment thing that I'm doing. 00:19:40.080 --> 00:19:44.040 And it's like a retirement planner, which is fine, you know, but it's not invested in like 00:19:44.090 --> 00:19:45.520 the same sense that you all are. 00:19:45.840 --> 00:19:49.520 And the AI like gives you recommendations on how your retirement plan is going. 00:19:49.940 --> 00:19:51.640 And I get the sense, I don't know for sure. 00:19:51.720 --> 00:19:52.160 I haven't used enough. 00:19:52.360 --> 00:19:56.920 I get the sense they're using a cheap model to get fast results, which makes me sad. 00:19:57.040 --> 00:20:02.560 Like if they were using Opus or GPD5 Pro, you could get such good answers. 00:20:03.080 --> 00:20:07.640 But it's really, really quick and responding just makes me real nervous that the answers are super shallow. 00:20:07.820 --> 00:20:10.900 So how do you, I mean, it's important the answers and the details you give to people. 00:20:11.000 --> 00:20:13.360 They're making really large decisions. 00:20:13.800 --> 00:20:16.500 It's not like, oh, I'll buy $1,000 of this stock. 00:20:16.900 --> 00:20:21.080 You know, I'll buy thousands of shares of this stock type of thing, given your clientele. 00:20:21.240 --> 00:20:24.060 So how do you make sure the AI is on rails? 00:20:24.500 --> 00:20:25.780 No, that's an excellent question. 00:20:26.080 --> 00:20:29.540 And to your point in finance, like trust is the name of the currency. 00:20:29.880 --> 00:20:35.120 And for our customers, it is essential that we are at 100% accuracy with every data point 00:20:35.160 --> 00:20:38.000 that we report because the decisions made are so consequential. 00:20:38.740 --> 00:20:42.520 To answer your question, there's quite a bit of discipline that we have within the Tetrix 00:20:42.760 --> 00:20:46.360 engineering processes to make sure that we are able to get more deterministic nature 00:20:46.480 --> 00:20:48.860 of outputs from something that's inherently probabilistic. 00:20:49.620 --> 00:20:54.480 And part of those, without disclosing too much of the secret sauce, but some top things are A, 00:20:54.840 --> 00:21:00.320 we invested a lot upfront in building our AI and evaluation harnesses. The good news is that the 00:21:00.480 --> 00:21:06.340 fact that we need 100% data accuracy means that we have very, very good data sets that are already 00:21:06.520 --> 00:21:11.400 annotated, have a human in the loop in terms of making sure it's 100%. And when we run experiments 00:21:11.700 --> 00:21:16.619 against those harnesses and those evals, it allows us to move really quickly and also figure out 00:21:16.640 --> 00:21:20.720 very quickly if something is breaking or not. So that was number one, because you can't solve what 00:21:20.720 --> 00:21:25.600 you don't know. Number two was most of our team has a lot of financial background, whether it is 00:21:25.650 --> 00:21:30.720 Grant, obviously, as an ex-investment banker, my co-founder or others. And what we have done is take 00:21:30.880 --> 00:21:36.740 that institutional knowledge of finance and encode it into a set of over 250 financial rules that are 00:21:36.920 --> 00:21:41.800 checking the outputs generated by the LLMs to make sure that things actually make sense from a 00:21:42.020 --> 00:21:45.299 financial perspective, whether it's within a document or when you look at things from a time 00:21:45.320 --> 00:21:50.200 series perspective. And number three is we have also built a lot of feedback loops so that whenever 00:21:50.340 --> 00:21:54.600 there's a human review or there's an erroneous data point spotted, we actually learn from that 00:21:54.840 --> 00:21:58.380 behavior and make sure that the subsequent extractions keep getting better and better. 00:21:58.780 --> 00:22:03.300 So with some of these things, we have been able to really get very high accuracy. And in private 00:22:03.480 --> 00:22:08.160 markets, I would argue we have the highest accuracy out of the box where our average accuracy is 96%. 00:22:09.020 --> 00:22:13.280 And that number keeps ticking upwards as we continue investing in this part of the platform. 00:22:13.840 --> 00:22:20.860 cool. So why Startup Row? How'd you end up there? Yeah, so we applied for Startup Row to, so we 00:22:20.940 --> 00:22:28.660 wanted to be more integrated into the Python community and also to expand our team. So we 00:22:28.980 --> 00:22:34.840 applied given that PyCon's the largest event conference for Python developers and our whole 00:22:35.240 --> 00:22:39.980 backend is built in Python. We wanted to kind of go there for a recruiting perspective, get the best 00:22:39.980 --> 00:22:44.760 and brightest talent and also see what other people, other companies and, you know, individuals 00:22:44.900 --> 00:22:49.680 are working on, whether it be open source projects or, you know, companies that we use 00:22:49.680 --> 00:22:50.740 on a day to day basis. 00:22:51.820 --> 00:22:51.900 Yeah. 00:22:51.940 --> 00:22:57.700 So basically, and also to tell people about what we're doing and kind of get the word 00:22:57.840 --> 00:22:59.100 out about Tetrix. 00:22:59.440 --> 00:22:59.860 Yeah, cool. 00:23:00.160 --> 00:23:03.480 So you're looking for possibly hiring some Python folks? 00:23:03.980 --> 00:23:04.060 Yep. 00:23:04.320 --> 00:23:09.040 Across the stack, we're looking to definitely scale up our engineering team. 00:23:09.200 --> 00:23:14.880 so yeah, I mean, front end, full stack back end, but you know, from a, an AML, 00:23:14.900 --> 00:23:18.820 but from a coding perspective, like definitely at least a lot of their backend data pipeline, 00:23:19.060 --> 00:23:20.600 that kind of thing, fully in Python. 00:23:21.060 --> 00:23:22.720 Well, I was talking to someone else. 00:23:22.840 --> 00:23:26.240 I won't call them out explicitly just now, but they had a really interesting insight 00:23:26.700 --> 00:23:30.300 saying that, you know, AI is going to change our jobs, obviously. 00:23:30.620 --> 00:23:35.439 I mean, that's not a huge revelation, but, but that it's probably going to shrink really 00:23:35.460 --> 00:23:40.680 large companies, because there's kind of a lot of support that maybe the AIs can do, but it's also 00:23:40.760 --> 00:23:46.100 going to expand out the existence of small companies, what small companies can do and so on. 00:23:46.140 --> 00:23:50.680 And it kind of sounds like you're a bit of an embodiment of that. Yeah, I think that's a fair 00:23:51.200 --> 00:23:56.440 analysis. I mean, we are a small but mighty team. We're growing, adding headcount as needed, 00:23:56.580 --> 00:24:02.559 but AI is obviously supercharging productivity internally as well. So what a person can deliver 00:24:02.860 --> 00:24:04.900 in a unit time has for sure increased 00:24:04.950 --> 00:24:06.460 and we want to keep leveraging those things 00:24:06.510 --> 00:24:07.380 to make sure that's true. 00:24:07.640 --> 00:24:08.600 But at the same time, 00:24:09.360 --> 00:24:11.260 there's so much demand for the product at the moment 00:24:11.400 --> 00:24:12.980 that capacity still is the constraint 00:24:13.200 --> 00:24:16.240 and hence why we're very excited to bring on more folks 00:24:16.360 --> 00:24:18.100 who have Python expertise into our community 00:24:18.500 --> 00:24:19.980 and into our team and work with them. 00:24:20.940 --> 00:24:21.940 That sounds really good. 00:24:21.990 --> 00:24:23.120 That sounds really positive. 00:24:23.270 --> 00:24:24.420 Like things are going well for you. 00:24:24.680 --> 00:24:26.680 I talked to Jason, the organizer of Startup Row 00:24:26.920 --> 00:24:28.660 and pointed out how many of the companies 00:24:28.770 --> 00:24:30.620 are either still going, got acquired 00:24:30.960 --> 00:24:32.460 or went public or whatever. 00:24:32.640 --> 00:24:34.080 So there's a good track record there. 00:24:34.340 --> 00:24:36.620 So how did your experience at Startup Row go? 00:24:37.000 --> 00:24:38.020 Was it worth going? 00:24:38.580 --> 00:24:41.840 Did you meet people who might be investors or partners? 00:24:42.220 --> 00:24:44.180 Did you find people who might build some of these roles? 00:24:44.600 --> 00:24:45.300 Was it worth going? 00:24:45.600 --> 00:24:46.180 Yeah, definitely. 00:24:46.440 --> 00:24:48.620 I mean, it was definitely a really great experience. 00:24:48.760 --> 00:24:52.660 I mean, we got to basically talk with a lot of people from different backgrounds. 00:24:53.920 --> 00:24:56.540 Yeah, I found, I mean, definitely a lot of promising candidates. 00:24:56.680 --> 00:25:08.660 And also it's very just cool to talk about our company to people who are like kind of like very large established financial services companies and tell them about what we're doing. 00:25:09.620 --> 00:25:11.480 They kind of get the business problem. 00:25:11.880 --> 00:25:23.840 But it's also really cool to see that we're in kind of the same like we're in Startup Row, but like we're talking to the same people and kind of being involved in the Python community as some of these really large established players. 00:25:24.200 --> 00:25:30.360 Yeah, when you all describe what you're doing, I'm like, there's something with a long sales cycle. People don't go to your site and just put in the credit card. 00:25:31.540 --> 00:25:31.900 Very true. 00:25:33.740 --> 00:25:37.880 The laugh of, you have no idea how hard this is to sell to enterprise, right? 00:25:39.480 --> 00:25:47.780 Just the timeframe from first contact to like, yeah, we'll do it is I'm sure really, really large. I've been on that site a little bit and it's kind of mind blowing. 00:25:48.160 --> 00:25:50.080 All right, a couple of things to wrap things up here. 00:25:50.550 --> 00:25:51.980 Give us a look inside, 00:25:52.700 --> 00:25:54.660 like a tech architectural look 00:25:54.660 --> 00:25:55.540 of what you're all building. 00:25:55.760 --> 00:25:57.800 You know, it's got some Python stuff going on, 00:25:58.260 --> 00:26:00.160 but what kind of tech is at play 00:26:00.360 --> 00:26:01.380 without giving away secret sauce? 00:26:01.800 --> 00:26:04.740 Yeah, so generally like our whole backend is Python. 00:26:05.120 --> 00:26:09.600 So we use FastAPI for all our, you know, HTTP services, 00:26:10.480 --> 00:26:13.200 use Pydantic for models and data validation, 00:26:14.960 --> 00:26:16.800 use a lot of like, you know, 00:26:16.860 --> 00:26:22.980 when we use AI, we're usually calling or doing things from a Python client for all of our data 00:26:23.420 --> 00:26:29.620 pipelines. Yeah. And I mean, we obviously use like other technologies, but, you know, I mean, 00:26:29.840 --> 00:26:34.160 be it like, you know, pandas or, you know, NumPy for like kind of more of the analytic side and then 00:26:34.900 --> 00:26:42.100 different kinds of packages for doing OCR and data extraction and all that stuff is basically Python. 00:26:42.660 --> 00:26:58.520 Fun. It's a cool tech stack. It's really neat to work with these things these days. It's a lot of nice tools. So what's on the roadmap? You are host startup row. You've had a lot of conversations. Where are you leading? Maybe did also did some of these conversations influence you a bit on what you should focus on? 00:26:58.700 --> 00:27:28.560 That's a good question. So, well, we have some exciting fundraising announcements coming soon. And as part of that, we obviously are going to spend a lot of time doubling down on the product and delivering the best customer experience possible. So in terms of some things that we're excited about, and I'll put the business hat on and then the tech nerd hat on. But on the business hat side, we're working on some very interesting new features. So if you look at Tetrix, you can kind of think of it as a three layered cake. There's the data capability layer, there is the workflows layer, and then there's the inside. 00:27:28.620 --> 00:27:33.140 layer that we deliver to customers. The data layer, we've been spending a lot of time since 00:27:33.580 --> 00:27:36.800 the start of the company. And I think we're in a really, really good position there. 00:27:37.300 --> 00:27:41.520 And we really want to start double clicking on the workflow and the insights there. So a lot of our 00:27:41.640 --> 00:27:45.820 time is being spent for use cases specific to our customer on those two layers, whether it is 00:27:46.080 --> 00:27:50.820 being able to diligence, net new opportunities faster, whether it's being able to get flexible 00:27:51.100 --> 00:27:55.719 outputs in terms of investor memos and some of those things. So we'll be doubling down on those 00:27:55.740 --> 00:28:00.240 things. On the tech side, where we're spending a lot of time on the roadmap is scaling and 00:28:00.400 --> 00:28:05.380 optimization. As we grow as a company, the amount of data points we are seeing is way larger than 00:28:05.380 --> 00:28:09.760 we have in the past. And that obviously puts pressure on the systems itself, but is a very 00:28:10.000 --> 00:28:14.760 fun and interesting technical problem of how do you end up scaling that data. AI is moving super 00:28:14.980 --> 00:28:18.800 quickly. So all of us have the pulse on what's going on. And we're always thinking of ways to 00:28:18.960 --> 00:28:23.360 kind of leverage that if it makes sense within our product. So those are some of the many, 00:28:23.500 --> 00:28:27.200 many, many highlights that we have on the roadmap right now, but hopefully gives a good teaser of 00:28:27.300 --> 00:28:29.980 how we think through the product and how we think throughout on the tech side. 00:28:30.240 --> 00:28:36.820 Cool. That sounds very great. Very nice plans there. So final thought, as we wrap up our time 00:28:37.040 --> 00:28:42.040 together, what advice would you have for people who maybe are thinking to start a company or have 00:28:42.240 --> 00:28:47.900 just started or maybe pre-revenue or something along those lines who want to be on startup row? 00:28:48.260 --> 00:28:52.920 I guess we each can take this one. I'm happy to start us off. I think general advice for a startup, 00:28:53.120 --> 00:28:54.740 don't be a solution in search of a problem. 00:28:55.120 --> 00:28:56.640 It should be the flipped way around. 00:28:56.840 --> 00:28:58.780 So put in all the effort upfront 00:28:58.880 --> 00:29:00.460 in terms of really understanding your customers, 00:29:00.640 --> 00:29:01.620 the willingness to pay, 00:29:02.000 --> 00:29:03.020 what value are you providing 00:29:03.460 --> 00:29:04.500 relative to what's out there? 00:29:04.580 --> 00:29:05.680 What's your unique point of view? 00:29:06.060 --> 00:29:06.480 Those things. 00:29:06.620 --> 00:29:08.580 So that's the general word of advice there. 00:29:08.920 --> 00:29:10.540 And I think for startup role specifically, 00:29:10.780 --> 00:29:12.300 the advice is be very intentional 00:29:12.300 --> 00:29:13.880 of why you're showing up to PyCon. 00:29:14.440 --> 00:29:16.160 Everyone's time is precious, including yours. 00:29:16.240 --> 00:29:18.480 So make sure there is a true objective there. 00:29:18.960 --> 00:29:20.880 And one thing I didn't realize actually going in 00:29:20.940 --> 00:29:21.960 and wish I had known earlier, 00:29:22.180 --> 00:29:23.820 Lightning talks fill up really fast. 00:29:24.780 --> 00:29:28.320 So more tactically, make sure you sign up for lightning talks ahead of time. 00:29:29.100 --> 00:29:32.460 So that's my piece of advice generally and with regards to PyCon. 00:29:32.760 --> 00:29:33.320 Excellent, Grant. 00:29:33.640 --> 00:29:38.720 Yeah, so for me, I was at a startup prior to Tetrix, which did not succeed. 00:29:39.540 --> 00:29:42.680 And being part of Tetrix, which is doing really well. 00:29:42.920 --> 00:29:48.359 I think the key thing is having really concrete problems that you can solve for your customers 00:29:48.740 --> 00:29:51.120 and being able to, you know, 00:29:51.470 --> 00:29:54.040 be able to have a product that really alleviates 00:29:54.360 --> 00:29:56.660 a pain point that people are willing to pay money for 00:29:56.830 --> 00:29:58.740 versus like, this isn't like nice data to have, 00:29:58.960 --> 00:30:00.120 nice thing to do, 00:30:00.340 --> 00:30:01.500 but versus something that's like, 00:30:01.800 --> 00:30:04.760 wow, this really like revolutionizes the way I do my job 00:30:05.000 --> 00:30:05.980 or anything like that. 00:30:06.340 --> 00:30:08.500 And the fact that we've been able to, you know, 00:30:09.460 --> 00:30:11.340 you know, really scale up and gain like 00:30:11.660 --> 00:30:14.000 these really high profile customers 00:30:14.320 --> 00:30:16.439 and we're kind of like transforming their whole experience 00:30:16.460 --> 00:30:18.820 just means that we're solving a really important problem. 00:30:19.300 --> 00:30:20.740 And that's kind of like, you know, 00:30:20.750 --> 00:30:24.200 what I've seen at, you know, at Tetrix versus my prior company. 00:30:24.450 --> 00:30:26.060 And I'd say for startups coming to Startup Row, 00:30:26.340 --> 00:30:31.040 I mean, I think just be really like curious 00:30:31.280 --> 00:30:33.100 and open to different people and like, you know, 00:30:33.400 --> 00:30:36.260 see what, really be interested in what people are working on. 00:30:36.520 --> 00:30:38.980 You know, maybe someone comes, talks to you, 00:30:39.140 --> 00:30:40.480 they have no idea what you're doing, 00:30:40.860 --> 00:30:42.280 but you can at least like go and explain 00:30:42.510 --> 00:30:43.480 like how your company's working, 00:30:43.640 --> 00:30:44.180 how you're using Python, 00:30:44.440 --> 00:30:46.100 kind of connect from an engineering perspective. 00:30:46.580 --> 00:30:48.180 because a lot of those like engineering problems 00:30:48.360 --> 00:30:49.160 are kind of the same. 00:30:49.460 --> 00:30:51.760 I mean, it's all, you know, data and users 00:30:52.060 --> 00:30:54.440 and, you know, dealing with those types of problems. 00:30:54.740 --> 00:30:57.200 So just trying to educate, 00:30:57.460 --> 00:30:58.940 but also take in as much as you can. 00:30:59.060 --> 00:30:59.280 Cool. 00:30:59.640 --> 00:31:01.180 Well, let me just riff, 00:31:01.380 --> 00:31:03.080 just a moment before we close it out here. 00:31:03.520 --> 00:31:04.820 I've always said that it's, 00:31:05.480 --> 00:31:07.560 I know a lot of people are getting into tech 00:31:07.620 --> 00:31:08.460 or getting into programming. 00:31:08.520 --> 00:31:10.420 They're like, I want to be a software developer 00:31:11.480 --> 00:31:12.380 and I'm choosing APIs 00:31:12.660 --> 00:31:15.060 or I'm choosing, you know, data science or something. 00:31:15.120 --> 00:31:15.940 And that's great. 00:31:16.240 --> 00:31:24.940 But I think the real secret sauce of being super successful is having some specialty like investment banking and some programming skill. 00:31:25.330 --> 00:31:32.020 And if you're quite good at both and you intersect them, you're all of a sudden extremely powerful in a very, very small space, right? 00:31:32.180 --> 00:31:43.900 Like taking your knowledge of banking and turning it into this product with programming and AI skills puts it into a really rare space, which I think that's a really good way to lever up your career. 00:31:44.100 --> 00:31:46.740 as people are thinking like, oh, it's hard to get a job at a big tech company. 00:31:46.820 --> 00:31:50.760 It's like, well, maybe you don't check just chase just straight down the line of I'm just 00:31:50.880 --> 00:31:54.020 going to be a web developer, but some kind of intersection with your specialty or your 00:31:54.210 --> 00:31:54.300 passion. 00:31:54.580 --> 00:31:54.940 Yeah, exactly. 00:31:55.150 --> 00:31:58.740 And one of the great things about startup is like you're really close to the business 00:31:58.990 --> 00:32:03.320 problem and you see kind of like software and it's more targeted how you're going to 00:32:03.470 --> 00:32:07.900 solve like actual problems that your users are going to are facing and you're getting 00:32:07.900 --> 00:32:10.480 the feedback live and it's really rewarding environment. 00:32:11.420 --> 00:32:16.980 And also you get to learn a ton because you're basically working across the stack on every 00:32:17.160 --> 00:32:18.300 problem that you're getting. 00:32:18.660 --> 00:32:21.380 And it's really an awesome way to learn programming. 00:32:21.820 --> 00:32:22.320 Yeah, that's awesome. 00:32:22.380 --> 00:32:23.080 I love small companies. 00:32:23.520 --> 00:32:26.000 Now, Ned, you got the final comment for the show. 00:32:26.380 --> 00:32:27.160 Let's wrap up the segment. 00:32:27.500 --> 00:32:31.600 It's just the thing that we were saying reminded me of back in the day, there was this analogy 00:32:31.660 --> 00:32:35.680 of being a T-shaped developer where you cut across the stack and are deep in one thing 00:32:35.880 --> 00:32:36.360 technology-wise. 00:32:36.460 --> 00:32:38.780 I think now it's almost like you have two T's to what you say. 00:32:38.820 --> 00:32:42.180 You have like your technology T, but you also have your domain expertise T. 00:32:42.380 --> 00:32:48.300 And that's kind of the next metamorphosis, I guess, of how the engineering role is going to play out. 00:32:48.940 --> 00:32:51.460 But my final word is just thank you again to the Python community. 00:32:51.640 --> 00:32:53.020 Thank you to you, Michael, for having us. 00:32:53.780 --> 00:32:57.520 And we're really excited as a company at Tetrix here to be deepening our partnerships. 00:32:58.160 --> 00:33:03.700 And if anyone is looking for roles or wants to join an exciting startup, Tetrix is always on the lookout for the best and the brightest. 00:33:04.460 --> 00:33:06.800 And we'd love to continue the conversation as we continue to scale. 00:33:06.980 --> 00:33:08.300 Awesome. Well, congratulations. 00:33:08.640 --> 00:33:13.400 you guys for your success so far and wishing you more in the future talk to you later thank you so 00:33:13.410 --> 00:33:22.140 much michael talk to you thanks michael next up we have david from arcjet david welcome to talk 00:33:22.380 --> 00:33:27.160 python to me awesome to have you here for this startup row segment yeah thanks for having me 00:33:27.370 --> 00:33:31.780 it's gonna be fun it was nice to meet you at long beach i enjoyed the conference a lot how was it for 00:33:31.790 --> 00:33:36.419 you yeah it was good and the python community is great so always good to be there in person it's 00:33:36.440 --> 00:33:42.460 not my first one. So it's always good. It's a little rejuvenating, at least for me, to 00:33:42.860 --> 00:33:46.720 actually connect with the humans on the other side of the docs and projects and so on. 00:33:46.900 --> 00:33:51.220 Yeah, absolutely. Yeah, well, let's just start with who you are. Do a quick introduction for 00:33:51.280 --> 00:33:57.180 the listeners. Yeah, I'm David Mitton. I'm the founder and CEO of Artjet, which is a runtime 00:33:57.740 --> 00:34:05.459 application security startup. I also run a DevTools newsletter called Console.dev, where I review 00:34:05.480 --> 00:34:06.440 interesting developer tools. 00:34:07.400 --> 00:34:10.419 35,000 subscribers have been doing that for almost six years now. 00:34:10.720 --> 00:34:14.419 But the day job is helping developers secure their applications. 00:34:14.980 --> 00:34:15.379 Very cool. 00:34:15.700 --> 00:34:18.639 So I'll be sure to put your newsletter in the show notes. 00:34:18.800 --> 00:34:19.520 People can check it out. 00:34:19.740 --> 00:34:22.159 But security, you sure it's not just a fad? 00:34:23.320 --> 00:34:24.220 Yeah, everything is secure. 00:34:24.270 --> 00:34:24.899 No problems. 00:34:25.179 --> 00:34:26.000 We've solved security. 00:34:26.220 --> 00:34:26.879 You can all go home. 00:34:27.120 --> 00:34:27.520 Yeah, exactly. 00:34:28.060 --> 00:34:32.419 We have so many tools and we have so much more awareness, but we also have a lot more 00:34:32.440 --> 00:34:35.480 clever people trying to do bad things, right? 00:34:35.570 --> 00:34:37.940 Like just look at all the supply chain issues we're running into. 00:34:38.240 --> 00:34:40.860 Right. Supply chain is a real problem. 00:34:41.280 --> 00:34:43.080 And it's been ignored for a very long time. 00:34:43.500 --> 00:34:46.560 There's some really interesting tools that are coming up around that. 00:34:47.320 --> 00:34:50.679 Artjet sits in your application once you deploy into production. 00:34:50.909 --> 00:34:53.820 So we protect against attacks coming in, automation. 00:34:54.370 --> 00:34:57.760 We help you enforce budget controls, deal with prompt injection. 00:34:58.620 --> 00:35:01.580 But the supply chain is just as important. 00:35:02.100 --> 00:35:03.960 I kind of split it pre-prod and prod. 00:35:04.240 --> 00:35:06.040 And we're probably on to year end production. 00:35:06.280 --> 00:35:08.340 And then pre-prod is all the code scanning, 00:35:08.860 --> 00:35:10.620 dependency management, all those kinds of things. 00:35:10.940 --> 00:35:11.120 Yeah. 00:35:11.440 --> 00:35:13.800 Some of the worst attacks have actually been 00:35:13.820 --> 00:35:15.440 through this whole supply chain side. 00:35:15.520 --> 00:35:18.960 You probably remember long ago 00:35:19.160 --> 00:35:20.520 when the internet first came into existence, 00:35:21.080 --> 00:35:23.060 some of these systems didn't have passwords. 00:35:23.260 --> 00:35:24.040 It's just, what's your username? 00:35:24.460 --> 00:35:25.140 Oh, okay. 00:35:25.280 --> 00:35:27.360 Hi, welcome back, Dr. Falcon or whatever. 00:35:28.140 --> 00:35:29.780 And it seems so naive. 00:35:29.980 --> 00:35:30.740 Like, how could that be? 00:35:30.880 --> 00:35:36.460 But I think, you know, pip install something or npm install, it's almost, it's just inviting 00:35:36.630 --> 00:35:40.240 other people's code that you don't know to run on your computer without checks, at least 00:35:40.380 --> 00:35:41.240 traditionally, which is insane. 00:35:41.800 --> 00:35:45.500 So I don't want to go down that rat hole, but it's just, you know, the thought of that 00:35:45.840 --> 00:35:48.040 analogy there is like a little stark, I think. 00:35:48.410 --> 00:35:49.540 But let's talk about ArcGIS. 00:35:49.850 --> 00:35:53.480 So it is security that runs inside your code. 00:35:53.870 --> 00:35:56.920 Tell us, I know you gave a little bit of a hint there, but tell us a little bit more about 00:35:56.970 --> 00:35:57.060 it. 00:35:57.360 --> 00:36:03.880 developers have this really bad reputation for not caring about security and that was front of mind 00:36:04.260 --> 00:36:08.960 when i started the company because like you said there are so many security tools it's just they're 00:36:08.960 --> 00:36:14.180 not built for developers they're outside of your code they're outside of your editor they don't have 00:36:14.260 --> 00:36:20.200 any connection to what's going on inside your code base and so developers would often just delegate 00:36:20.500 --> 00:36:27.320 that to a third-party service or a different team and just like with devops where developers take 00:36:27.320 --> 00:36:33.000 responsibility for the applications that they're building, I think we can apply similar principles 00:36:33.360 --> 00:36:36.820 to security. Whereas if you give developers the right tools and the right principles, 00:36:37.470 --> 00:36:41.080 they can just think of security as another feature that they have to think about and build. 00:36:41.600 --> 00:36:45.920 And so that's what Artjet is. It's an SDK. You drop it into your application dependencies. 00:36:46.800 --> 00:36:53.000 We support Python, JavaScript, TypeScript, and Go. And then it gives you these building blocks to 00:36:53.020 --> 00:36:59.320 solve problems and the problem framing is critical because developers when they have to think about 00:36:59.540 --> 00:37:05.700 security often do it once as a problem they're not buying a waft they're not trying to buy code 00:37:05.900 --> 00:37:11.020 scanning static analysis tools they don't search for those things they search for i'm getting spam 00:37:11.240 --> 00:37:18.860 signups or someone's abusing my application and i need to apply budget controls these are specific 00:37:18.880 --> 00:37:20.100 developer-focused problems. 00:37:20.550 --> 00:37:22.900 And that's where we try and help developers. 00:37:23.010 --> 00:37:25.260 We give them building blocks for bot detection, 00:37:25.920 --> 00:37:28.520 for sign-up spam protection, for prompt injection. 00:37:29.080 --> 00:37:32.280 And more recently, we're creating these in a way 00:37:32.480 --> 00:37:34.140 that means you can just get your coding agent 00:37:34.640 --> 00:37:35.780 to solve it for you. 00:37:35.960 --> 00:37:37.040 That's what you've got on the screen there. 00:37:37.500 --> 00:37:39.380 And you just install our skill, 00:37:39.600 --> 00:37:41.000 and then you ask your coding agent, 00:37:41.460 --> 00:37:43.280 set a bar jet, and it will go through, 00:37:43.740 --> 00:37:46.500 examine your code, figure out what endpoints, 00:37:46.600 --> 00:37:51.560 what routes, what tool calls need protection, and then install OutJet and help you solve the 00:37:51.720 --> 00:37:57.640 security problem. That is really neat. I love how it knows, basically bootstraps itself onto the 00:37:57.770 --> 00:38:03.100 system by teaching the AI agents how to do things. I'm really starting to notice people deeply 00:38:03.340 --> 00:38:10.500 thinking about how to support AI agents that work alongside developers for their projects. I just 00:38:10.500 --> 00:38:16.480 had the great tables guys from Positon, and they've got a whole section on agents and skills and 00:38:16.500 --> 00:38:22.260 stuff so and CLI so that if you want to create documentation it's not just you create it but 00:38:22.580 --> 00:38:26.060 here's the tools that the agents can use to work this to help you create it as well and it sounds 00:38:26.060 --> 00:38:31.560 like you got something real similar we started with developer experience because the insight 00:38:32.040 --> 00:38:37.720 I had was that the best developer tool this is from from reviewing thousands of tools with console.dev 00:38:38.040 --> 00:38:42.580 but the best developer tools have the best developer experience so that means good documentation 00:38:42.600 --> 00:38:50.600 I mean, it means proper comments that your editor can surface as the IntelliSense type 00:38:51.420 --> 00:38:54.580 pop-ups that show up in your editor to give you hints by what you should implement, 00:38:55.100 --> 00:38:55.960 creating examples. 00:38:56.720 --> 00:39:03.420 And we started in 2023 before coding agents really became as dominant as they are now. 00:39:03.960 --> 00:39:09.060 But what we created by accident was this idea of agent experience, because good developer 00:39:09.080 --> 00:39:15.340 experience is very similar to what you need to do to have good agent experience and so today that 00:39:15.460 --> 00:39:21.100 means we have skills you can install we've got an mcp server so you can query your data or your agent 00:39:21.130 --> 00:39:27.500 can query your data and to help you understand what you need to do we've got a cli and we make 00:39:27.500 --> 00:39:35.540 it super easy for you to just point your coding agent at our docs or our lms.txt and then it will 00:39:35.520 --> 00:39:36.840 figure out what needs to be done. 00:39:37.040 --> 00:39:40.960 Like if you scroll up a little bit, the CTA that we have on the homepage, the 00:39:41.100 --> 00:39:45.200 first one is start with a prompt and you can go to the code if you want, but 00:39:45.340 --> 00:39:49.140 install the skill and then you tell your agent what to do. 00:39:49.600 --> 00:39:53.500 And where developers would often get stuck is you've done that initial 00:39:53.980 --> 00:39:55.360 installation or then what? 00:39:55.560 --> 00:39:59.260 Because that's in your code, you have all the context of who the user is and 00:39:59.360 --> 00:40:01.080 you can log rather than blocking. 00:40:01.660 --> 00:40:05.000 And when you deploy, your agent can follow along, can follow the logs. 00:40:05.180 --> 00:40:06.340 It can connect to Archer. 00:40:06.390 --> 00:40:07.620 It can pull all the data out. 00:40:08.300 --> 00:40:12.200 And then it can make suggestions to change the rules to avoid false positives and then 00:40:12.440 --> 00:40:13.220 commit them for you. 00:40:13.350 --> 00:40:18.260 And so what you get is this security partner that works with the Archer building blocks 00:40:18.390 --> 00:40:19.900 and helps you solve these security problems. 00:40:20.440 --> 00:40:20.780 I love it. 00:40:20.920 --> 00:40:22.820 That's a really great onboarding experience. 00:40:23.250 --> 00:40:28.340 I feel like there's this split of a lot of people having great experiences with AI agents 00:40:28.370 --> 00:40:29.880 and a lot of people not having them. 00:40:30.340 --> 00:40:34.280 And I know there's a lot going on there, but a lot of people not having great experiences, 00:40:34.480 --> 00:40:40.720 I think there's just steps they're missing or techniques they're missing, or they're choosing a really, really free, simple model. 00:40:41.140 --> 00:40:45.880 But things like this, like you just automatically run a single command to install the skills. 00:40:46.220 --> 00:40:48.400 And now AI is already better with this. 00:40:48.480 --> 00:40:50.080 It's really, really cool that you've thought through it. 00:40:50.260 --> 00:40:53.440 Obviously, having the AI work within your system. 00:40:53.660 --> 00:41:01.920 So when I'm using your SDK, is it actually doing AI at runtime or is AI sort of around the more static side? 00:41:02.000 --> 00:41:07.360 We use AI for the prompt injection detection side of the product. 00:41:08.000 --> 00:41:12.500 And that happens when you build in that particular building block into your code. 00:41:12.680 --> 00:41:13.760 But we have other components. 00:41:14.400 --> 00:41:18.020 We do PII detection, so we can detect email addresses, credit card numbers. 00:41:18.620 --> 00:41:19.980 And that happens entirely locally. 00:41:20.400 --> 00:41:25.880 We actually ship a WebAssembly sandbox with the SDK, which runs in process in your environment. 00:41:25.940 --> 00:41:30.700 And so when someone submits their credit card number by accident into your support form, 00:41:30.970 --> 00:41:34.480 we do that analysis entirely locally and it never leaves your environment. 00:41:34.880 --> 00:41:36.840 But to do something like prompt injection detection, 00:41:37.260 --> 00:41:40.580 that's where we run a load of specialized models on our cloud. 00:41:41.200 --> 00:41:44.520 And so the call goes out to us and then we're running AI behind the scenes. 00:41:45.300 --> 00:41:48.860 The difference is just the kind of thing you want to protect against. 00:41:49.060 --> 00:41:51.860 Like if you're doing a form submission, then you want us to respond within, 00:41:52.920 --> 00:41:56.080 we set an SLA of 20 milliseconds to respond to those. 00:41:56.260 --> 00:41:57.240 So you would never notice. 00:41:57.550 --> 00:41:59.500 And if it's inside the WebAssembly sandbox, 00:41:59.800 --> 00:42:00.840 it's less than a millisecond. 00:42:01.180 --> 00:42:02.500 But if you're going to an LLM, 00:42:02.650 --> 00:42:04.900 then it may take a couple of seconds to get a response. 00:42:05.190 --> 00:42:06.600 And that's normal as expected. 00:42:06.980 --> 00:42:10.020 So we have a little bit of leeway to do our own analysis. 00:42:10.720 --> 00:42:12.320 Typically, that's around 200 milliseconds. 00:42:12.800 --> 00:42:16.120 Yeah, if you add half a second to someone else's LLM call, 00:42:16.480 --> 00:42:18.460 it probably, you couldn't tell the difference. 00:42:18.860 --> 00:42:19.700 It could just be, oh, well, 00:42:19.750 --> 00:42:21.380 the LLM's a little busy right now, right? 00:42:21.720 --> 00:42:22.660 Yeah, very cool. 00:42:22.860 --> 00:42:26.300 Okay, so why start this company? Why start ArcJet? 00:42:26.540 --> 00:42:32.800 I started it because I felt that the developer tooling for security just wasn't there. 00:42:33.300 --> 00:42:39.360 And we had gone through in the previous few years, kind of 2020 onwards, this revolution, 00:42:39.590 --> 00:42:43.280 certainly in the JavaScript ecosystem. And then it started going over into 00:42:44.120 --> 00:42:48.160 Python and into the other ecosystems as well, that you would get these amazing tools, 00:42:48.300 --> 00:42:49.680 You get amazing cloud platforms. 00:42:49.810 --> 00:42:53.640 You get amazing CLIs and tools for developers to write code. 00:42:53.840 --> 00:42:57.500 But it was five or 10 years behind in security. 00:42:57.870 --> 00:43:03.560 And I thought we could bring this new approach, the developer experience to security in your code. 00:43:03.960 --> 00:43:06.380 Something that's actually built for developers to try and solve problems. 00:43:06.860 --> 00:43:12.820 And I thought we could build out a platform company to solve all these different problems. 00:43:13.580 --> 00:43:15.420 But developers need these building blocks. 00:43:15.600 --> 00:43:18.520 And so that's why we've kind of split the product up into these different areas. 00:43:18.900 --> 00:43:22.120 So you can come to us for one thing, like prompt injection detection, 00:43:22.680 --> 00:43:27.620 but you probably also want rate limiting to implement budget controls for your AI tool. 00:43:28.260 --> 00:43:32.480 You want bot protection to stop automation attacks against your application. 00:43:32.560 --> 00:43:33.640 You want PAI detection. 00:43:34.060 --> 00:43:36.100 So all of these things start to compose together. 00:43:36.600 --> 00:43:42.460 You can use them individually or you can bring them into a multi-rule security policy. 00:43:43.140 --> 00:43:44.800 But that's the reason for creating this. 00:43:44.980 --> 00:43:46.260 Seems like a great idea to me. 00:43:46.720 --> 00:43:48.760 So Startup Pro, how'd you end up there? 00:43:49.500 --> 00:43:53.460 Yeah, I was following the Icon blog for a while 00:43:53.780 --> 00:43:56.500 because I knew we were going to do a Python SDK 00:43:57.780 --> 00:44:00.500 and wanted to start getting into the community. 00:44:01.000 --> 00:44:02.400 Python's actually the first programming language 00:44:02.940 --> 00:44:07.660 and I really learned and I've been writing it since 2008, 2009. 00:44:07.940 --> 00:44:09.820 So back in the Python 2 era. 00:44:10.520 --> 00:44:11.320 But we started out. 00:44:11.320 --> 00:44:12.280 The two to three wars. 00:44:13.140 --> 00:44:14.440 You're a veteran of the two to three wars. 00:44:14.600 --> 00:44:15.200 Yeah, I know. 00:44:15.900 --> 00:44:20.140 Luckily, I kind of stopped writing Python for a bit with my last startup. 00:44:20.740 --> 00:44:25.320 And then when I came back to writing code, it had already gone on to the 3, version 3. 00:44:25.480 --> 00:44:28.400 So I kind of skipped that pain. 00:44:29.780 --> 00:44:33.000 But we started with JavaScript and we just had a JavaScript SDK 00:44:33.300 --> 00:44:37.920 because all the new applications were being built using JavaScript and TypeScript. 00:44:38.660 --> 00:44:42.260 As AI started to come onto the scene properly 00:44:42.500 --> 00:44:45.300 kind of end of 2024 into 2025, 00:44:45.960 --> 00:44:47.980 Python really started to shine again 00:44:48.320 --> 00:44:51.520 because it's been native in the data community, 00:44:51.740 --> 00:44:53.120 the data science community for a long time. 00:44:53.520 --> 00:44:56.500 And it's where all the AI work is happening. 00:44:56.830 --> 00:45:00.320 And it started to be used for the AI backend APIs again. 00:45:00.700 --> 00:45:02.760 And we just built our own backend. 00:45:03.300 --> 00:45:04.980 I'm using the Open Inference Protocol 00:45:05.580 --> 00:45:07.879 and that allows us to run different models 00:45:07.940 --> 00:45:10.340 but have an abstracted layer internally 00:45:10.710 --> 00:45:13.980 because our actual core infrastructure is written in Go. 00:45:14.400 --> 00:45:16.100 That's what I moved to from Python to Go. 00:45:17.220 --> 00:45:19.800 But we've decided we were going to release the Python SDK, 00:45:20.240 --> 00:45:22.080 which came out at the beginning of this year. 00:45:22.900 --> 00:45:26.440 And that was just in time for the startup row application 00:45:27.200 --> 00:45:28.260 and to attend PyCon 00:45:28.540 --> 00:45:30.700 and start getting it in the hands of Python developers. 00:45:31.120 --> 00:45:31.360 Nice. 00:45:31.920 --> 00:45:34.060 Is that a little bit like Open Router and those other things, 00:45:34.520 --> 00:45:35.600 this middle layer? 00:45:36.080 --> 00:45:36.800 It's similar, yeah. 00:45:36.920 --> 00:45:40.700 The challenge with the models is they all require different things. 00:45:40.780 --> 00:45:41.460 They have different inputs. 00:45:42.240 --> 00:45:45.460 And if you're going to run multiple models, which is what we do, 00:45:45.560 --> 00:45:48.500 we kind of run the prompt injection detection in parallel 00:45:48.800 --> 00:45:50.880 across multiple models and then compare the results. 00:45:51.860 --> 00:45:54.460 And for our backend to have a consistent API 00:45:54.820 --> 00:45:57.760 where we get the result out, we want to get the token count, 00:45:58.340 --> 00:46:00.120 we want to monitor the latency, 00:46:00.640 --> 00:46:04.660 we created a thin layer that uses the open inference protocol, 00:46:05.120 --> 00:46:06.700 which is what our Go code connects to. 00:46:07.100 --> 00:46:10.140 And then that abstracts away the differences of the different models. 00:46:10.900 --> 00:46:15.100 And that makes it a lot easier to just build our internal client server architecture. 00:46:15.440 --> 00:46:15.880 Makes sense. 00:46:16.200 --> 00:46:17.720 So how did Startup Row go? 00:46:17.800 --> 00:46:18.780 Did you make some good connections? 00:46:19.560 --> 00:46:20.760 Did you feel like it was worth going? 00:46:21.080 --> 00:46:21.760 Yeah, it was great. 00:46:22.420 --> 00:46:29.000 And it allowed us to get in front of the community with the support of the PyCon organizers. 00:46:29.600 --> 00:46:35.180 It was interesting comparing to other conferences because to date, we've been mainly at JavaScript conferences. 00:46:35.940 --> 00:46:51.940 And the type of questions that we get really shows the understanding of building AI workflows, building AI APIs, building the data models and the data workflows behind the front end applications, which are still typically built in JavaScript. 00:46:52.540 --> 00:46:54.260 And just the questions are very different. 00:46:54.270 --> 00:47:00.240 And you could utilize a different type of engineer at PyCon than you do a JavaScript conference. 00:47:00.740 --> 00:47:02.460 Yeah, there's some really clever people there. 00:47:02.520 --> 00:47:07.480 I always feel like I'm talking to some really smart people in the room when I'm interacting with the PyCon attendees. 00:47:07.980 --> 00:47:11.740 So you talked a bit about the tech going on behind the scenes. 00:47:12.060 --> 00:47:14.500 It's interesting that you have Go, I think, there. 00:47:15.500 --> 00:47:21.620 It really seems like that's server-side type of programming first is what Go is for. 00:47:22.020 --> 00:47:29.140 I think people think Go Rust is kind of similar, but Rust is a little more lower-level operating system type of thinking, 00:47:29.300 --> 00:47:32.280 I think, whereas Go is all about async and programming. 00:47:33.160 --> 00:47:36.520 What are your thoughts? I know you're not doing it, so there's no reason to switch over. 00:47:36.520 --> 00:47:41.420 What are your thoughts about using some of the modern Python API things for your backend? 00:47:41.780 --> 00:47:47.960 We use FastAPI and we use an SGI app that's hosted on Modul for our AI inference. 00:47:48.860 --> 00:47:52.260 Modul gives us the GPUs and the hosting environment auto scales. 00:47:52.600 --> 00:47:55.720 And then our applications are just deployed in containers. 00:47:56.500 --> 00:48:04.800 The Go API, the Python SDK that our customers use is Go because it's a gRPC API. 00:48:05.440 --> 00:48:12.600 And Go has the best libraries available for gRPC because gRPC is written by Google and Go is written by Google. 00:48:12.680 --> 00:48:14.200 So they have the best APIs. 00:48:14.620 --> 00:48:16.380 I mean, yeah, they were built to go hand in hand. 00:48:16.440 --> 00:48:20.720 I feel like gRPC never really spread beyond Go. 00:48:21.080 --> 00:48:25.220 I mean, there's some options in Python to use it and some other languages, but it's kind of like, 00:48:25.500 --> 00:48:27.200 It still feels proof of concept to me. 00:48:27.680 --> 00:48:27.980 I don't know. 00:48:28.120 --> 00:48:29.480 I've looked at gRPC as well. 00:48:29.540 --> 00:48:32.680 I'm like, that'd be pretty neat, but no, just do JSON or something. 00:48:32.980 --> 00:48:34.540 Yeah, we did it because of the performance. 00:48:35.080 --> 00:48:36.900 So we talked a little bit about the latency. 00:48:37.460 --> 00:48:41.840 And if we're sitting in your request path, we want to respond as quickly as we possibly can. 00:48:42.200 --> 00:48:45.880 Like ignoring the prompt injection detection, which adds a couple of hundred milliseconds. 00:48:46.460 --> 00:48:51.500 When we're doing bot detection, when we're doing rate limiting, we want to give you a response within just a couple of milliseconds. 00:48:51.540 --> 00:48:59.280 And so when you install our Python SDK, it opens an HTTP2 socket that maintains persistent connection with our edge network. 00:48:59.730 --> 00:49:05.320 And that removes some of the latency because you don't have to establish a connection every time. 00:49:05.510 --> 00:49:12.280 But then we're making RPC calls using gRPC to minimize the latency because the wire protocol is binary. 00:49:12.690 --> 00:49:17.220 So it doesn't have to do the heavy work that you do when you're parsing JSON. 00:49:17.300 --> 00:49:22.960 and that just makes it a lot more performant and means that when the communication is happening we 00:49:23.100 --> 00:49:27.580 can we have backwards compatibility with the protocol layer it just gives us a lot of these 00:49:27.800 --> 00:49:33.140 nice properties and that allow us to have a stable but perform an api yeah it makes a lot of sense 00:49:33.580 --> 00:49:39.240 i'd probably use msgspec these days if i was really really worried about packet i think gRPC is 00:49:39.340 --> 00:49:43.839 also great but like if i was had to fit it into the python space i'd probably go message spec 00:49:44.020 --> 00:49:47.820 I think that's probably the tightest, most binary packing you can get these days. 00:49:48.220 --> 00:49:48.660 Yeah, definitely. 00:49:48.920 --> 00:49:53.500 There are some interesting options now, but we support multiple languages. 00:49:54.160 --> 00:49:55.580 So we started with JavaScript. 00:49:56.020 --> 00:49:59.000 There's some really good generators for gRPC in JavaScript. 00:50:00.300 --> 00:50:04.960 And we've just done a Go SDK as well, and we're going to go to other languages. 00:50:05.700 --> 00:50:11.800 So we looked at a few different options, and this was 2023, so it's a few years now. 00:50:12.460 --> 00:50:13.700 You had a lot of time to think about it. 00:50:13.800 --> 00:50:14.920 no one was going outside anymore. 00:50:15.920 --> 00:50:17.280 What a weird time. 00:50:17.680 --> 00:50:21.520 So give us, I want to maybe just get your perspective here 00:50:21.680 --> 00:50:23.320 since you've lived on both sides of these fences. 00:50:23.800 --> 00:50:25.980 How is JavaScript, TypeScript security 00:50:26.800 --> 00:50:28.360 different than Python right now? 00:50:28.600 --> 00:50:30.120 That is a good question. 00:50:30.480 --> 00:50:35.780 I think PyPI hasn't had the same problems as npm has. 00:50:36.100 --> 00:50:38.180 And it's probably a factor of its popularity 00:50:38.780 --> 00:50:41.000 with where all the attention is. 00:50:42.140 --> 00:50:43.039 That's going to change 00:50:43.360 --> 00:50:46.580 because these package registries are really set up 00:50:46.890 --> 00:50:48.140 almost as volunteer organizations. 00:50:49.020 --> 00:50:52.120 And certainly PyPI is run by the Python Foundation. 00:50:52.500 --> 00:50:54.520 So that is a kind of formal connection there. 00:50:55.640 --> 00:50:58.900 But they don't get the funding of a megacorp 00:50:59.120 --> 00:51:02.340 that should be able to invest huge amounts in security. 00:51:02.720 --> 00:51:04.500 Of course, npm is run by a megacorp. 00:51:04.510 --> 00:51:05.940 It's run by GitHub, which is Microsoft. 00:51:06.150 --> 00:51:09.020 And so they should have the resources to really improve security. 00:51:09.980 --> 00:51:11.999 But that just hasn't... 00:51:12.020 --> 00:51:13.680 Still, they're struggling under all the attacks. 00:51:13.980 --> 00:51:14.640 They're struggling. 00:51:14.860 --> 00:51:19.420 They just haven't allocated the resources it really deserves. 00:51:20.600 --> 00:51:22.040 And this is just a problem with the architecture 00:51:22.260 --> 00:51:23.460 because they were built in an era 00:51:23.700 --> 00:51:26.120 when this just wasn't an attack factor. 00:51:26.480 --> 00:51:29.260 It was just so much trust around open source. 00:51:29.440 --> 00:51:30.900 It's like, it's built for the good. 00:51:31.180 --> 00:51:32.360 They put it out there for you to use. 00:51:32.460 --> 00:51:34.260 It has a license, just install it. 00:51:34.920 --> 00:51:36.380 Well, maybe that's not the only problem. 00:51:37.080 --> 00:51:37.780 Yeah, that's right. 00:51:38.000 --> 00:51:39.580 It's just they provide a very basic service. 00:51:39.840 --> 00:51:42.080 It's basically just hosting your code in a certain way, 00:51:42.500 --> 00:51:43.380 but that's been abused. 00:51:43.700 --> 00:51:47.440 And it's multiple orders of magnitude increase in traffic. 00:51:48.140 --> 00:51:50.020 And that just means you're going to get bad actors 00:51:50.420 --> 00:51:51.480 and they're going to exploit it. 00:51:51.660 --> 00:51:51.900 Yeah. 00:51:52.100 --> 00:51:54.520 I honestly think that with all the agentic AI 00:51:54.740 --> 00:51:56.920 helping people write little bits of code here and there 00:51:57.940 --> 00:52:00.360 and stuff like the security supply chain thing 00:52:00.380 --> 00:52:00.940 that we're talking about, 00:52:01.080 --> 00:52:02.100 there's going to be a lot more vendoring. 00:52:02.320 --> 00:52:04.720 And if you only need one or two functions out of a library, 00:52:04.840 --> 00:52:06.260 maybe you could just have Cloud write it 00:52:06.460 --> 00:52:08.620 and you don't have to depend on anything and worry about it. 00:52:09.000 --> 00:52:13.880 yeah it's we'll see where it goes in the next five years agreed yeah that's why google's 00:52:13.930 --> 00:52:18.820 always run things they always vendor all their dependencies and now i think if it's just a basic 00:52:18.980 --> 00:52:23.060 dependency that's not really doing that much you can get your coding agent to do it for you 00:52:23.330 --> 00:52:28.220 i one of the things that i always talk to our customers about is that well you can write your 00:52:28.220 --> 00:52:32.739 own frame limiting library you can do that pretty easily and use redis but can you build your own 00:52:32.760 --> 00:52:39.280 bot detection and deal with all the kind of the arms race of keeping up with automation that's 00:52:39.280 --> 00:52:43.480 where we think we add value and i think that's the way we're going to see what dependencies is most 00:52:43.620 --> 00:52:47.800 of them you're going to write yourself or get your ai to do it but there's going to be a few key 00:52:47.960 --> 00:52:52.380 things that really you shouldn't be doing yourself like crypto cryptography not cryptocurrency 00:52:53.700 --> 00:52:57.140 that's going to be the that's going to be the questions like what is the crypto equivalent 00:52:57.700 --> 00:53:02.720 yep 100 if you see people doing 100 from scratch their own crypto that's pretty much a 00:53:02.740 --> 00:53:06.500 very big one, a red flag. So speaking of, you know, what people are going to be doing, 00:53:07.040 --> 00:53:11.180 what's the roadmap? Let's close it out with the roadmap for ArcGET. Where are things going from 00:53:11.200 --> 00:53:15.960 what you have now? We're going to add more language support. And it's not just a case of 00:53:16.280 --> 00:53:21.420 code genning a library because we have the WebAssembly component in there that does a lot of 00:53:21.520 --> 00:53:27.020 the security analysis in your environment. And that requires a bit of engineering work for each 00:53:27.040 --> 00:53:33.300 platform. In JavaScript, WebAssembly is native within Node, within the runtime, so we have to 00:53:33.610 --> 00:53:38.440 write the bindings, but it's relatively straightforward. In Python, we use WASM time, 00:53:38.570 --> 00:53:46.840 which is an open source runtime for Python. And in Go, we use one called WASERO, but different 00:53:47.520 --> 00:53:52.680 ecosystems have different packages because often WebAssembly is not natively built into the runtime. 00:53:53.300 --> 00:53:59.760 And so that's where a lot of the work goes, is making sure that WebAssembly works in a way that is performant and native. 00:54:00.340 --> 00:54:02.220 So we're going to add more language support. 00:54:02.700 --> 00:54:03.300 That's the main thing. 00:54:03.600 --> 00:54:03.780 Okay. 00:54:04.100 --> 00:54:06.280 Yeah, that's, I mean, got a solid foundation. 00:54:06.500 --> 00:54:11.260 Now you just want to make it available to all the different users and developers of different platforms. 00:54:12.380 --> 00:54:14.080 You know, it would be pretty interesting. 00:54:14.600 --> 00:54:21.340 In Python, we've got the CFFI and the C bindings and all these different interop things and Rust through PyO3. 00:54:21.680 --> 00:54:26.880 It'd be kind of interesting to just have a native execution of Wasm in the runtime without 00:54:26.980 --> 00:54:28.700 some kind of adaptive layer type of thing. 00:54:28.700 --> 00:54:29.440 I think that would be cool. 00:54:29.620 --> 00:54:34.560 Definitely, because I want to have absolutely nothing to do with C in a security focused 00:54:35.000 --> 00:54:35.140 application. 00:54:35.640 --> 00:54:37.540 I can imagine that. 00:54:37.800 --> 00:54:38.360 Yeah, definitely. 00:54:38.520 --> 00:54:43.420 It should be built into the runtime, but then we had the problem of minimum Python versions 00:54:43.700 --> 00:54:44.620 and all those kinds of things. 00:54:44.800 --> 00:54:50.380 And I think Python 3 people do typically stay more up to date, but it's still a problem when 00:54:50.400 --> 00:54:53.780 you require people to be on the very latest version of the runtime. 00:54:54.400 --> 00:54:55.080 Yeah, 100%. 00:54:55.080 --> 00:54:59.040 A lot of people just take what comes with Ubuntu or something, which is usually a year behind. 00:54:59.400 --> 00:55:02.920 All right, David, I appreciate you taking this time to chat with us. 00:55:03.060 --> 00:55:05.680 Congratulations on ArcJet and good luck for the future. 00:55:06.000 --> 00:55:06.180 Great. 00:55:06.520 --> 00:55:06.740 Thanks a lot. 00:55:07.080 --> 00:55:07.260 Yeah. 00:55:07.490 --> 00:55:07.620 Bye. 00:55:09.880 --> 00:55:12.480 Our next startup is femoral.dev. 00:55:13.090 --> 00:55:14.660 And we're going to be talking to Chinmaya. 00:55:16.400 --> 00:55:17.780 Chinmaya, welcome to Talk Python. 00:55:18.060 --> 00:55:18.720 Excellent to have you here. 00:55:18.980 --> 00:55:20.080 Hey, super nice to be here. 00:55:20.260 --> 00:55:23.420 Fun times hanging out in PyCon, meeting up on Startup Row. 00:55:23.940 --> 00:55:25.520 That was a really cool setup there. 00:55:26.000 --> 00:55:26.560 Yeah, it was really good. 00:55:26.560 --> 00:55:28.740 It was a pretty awesome opportunity to be there. 00:55:28.960 --> 00:55:31.040 So glad we made that happen. 00:55:31.340 --> 00:55:31.960 I can imagine. 00:55:32.000 --> 00:55:37.040 I was talking to one of the organizers and the success rate of startups on Startup Row 00:55:37.160 --> 00:55:37.620 is really high. 00:55:38.380 --> 00:55:38.960 Hoping to follow suit. 00:55:38.960 --> 00:55:39.620 I hope for it. 00:55:40.060 --> 00:55:40.300 Exactly. 00:55:41.120 --> 00:55:42.140 Let's keep it rolling, right? 00:55:42.420 --> 00:55:46.020 Well, let's actually begin by having you do a quick introduction for the listeners. 00:55:46.500 --> 00:55:46.860 Sure. 00:55:47.300 --> 00:55:47.360 Yeah. 00:55:48.240 --> 00:55:48.720 I'm Chinmaya. 00:55:49.000 --> 00:55:50.460 I'm the founder of Femeral. 00:55:50.780 --> 00:55:55.020 We're like a hosting platform for Python-based web apps. 00:55:55.620 --> 00:55:58.240 And yeah, just excited to be here. 00:55:58.640 --> 00:55:58.960 Awesome. 00:55:59.320 --> 00:56:02.440 I think hosting's been around forever, right? 00:56:02.540 --> 00:56:04.640 Like cPanel land and all that weird stuff. 00:56:04.880 --> 00:56:09.400 But it seems like there's a lot of opportunity to create better opportunities, better developer 00:56:09.680 --> 00:56:12.120 user experiences, and also cost, right? 00:56:12.400 --> 00:56:15.380 Yeah, that's kind of our angle, right? 00:56:15.460 --> 00:56:18.100 is like, obviously you can go and spin up a server. 00:56:18.290 --> 00:56:21.020 Pretty much every company under the sun offers that service. 00:56:22.040 --> 00:56:24.500 But, you know, we really want to focus on the developer experience, 00:56:25.120 --> 00:56:28.560 making it as seamless as possible to actually get your code 00:56:28.890 --> 00:56:30.300 hosted into a place where it's live 00:56:30.700 --> 00:56:34.980 and without having to, you know, deal with the manual steps 00:56:35.580 --> 00:56:38.300 and extra problem surface area that come with, 00:56:38.530 --> 00:56:39.920 you know, potentially managing your own hosting. 00:56:41.040 --> 00:56:44.460 And like you mentioned, the cost, that's also a factor for us 00:56:44.480 --> 00:56:48.180 because, you know, especially when you have persistently provisioned compute, you really 00:56:48.340 --> 00:56:54.140 have to size and scale your servers such that they don't, you know, break your wallet, break 00:56:54.150 --> 00:56:54.660 your bank account. 00:56:55.900 --> 00:56:58.300 And so you're kind of dealing with two problems, right? 00:56:58.460 --> 00:57:02.640 It's like, is my server or set of servers actually going to be able to handle all of this traffic? 00:57:03.100 --> 00:57:05.900 And at the same time, am I going to be able to pay for all of this? 00:57:06.820 --> 00:57:12.640 Yeah, that's both the promise and the potential downfall of the cloud is it's easy to set up 00:57:12.680 --> 00:57:13.040 a few things. 00:57:13.140 --> 00:57:15.500 they won't cost very much unless they take off and then maybe they do. 00:57:15.850 --> 00:57:15.940 Yeah. 00:57:16.070 --> 00:57:20.080 And then, you know, you're spending all this dev time, you know, trying to manage all that 00:57:20.280 --> 00:57:21.220 infra on top of that. 00:57:21.300 --> 00:57:25.800 So, you know, if you equate, you know, potentially your own developer time or your team's developer 00:57:26.000 --> 00:57:29.280 time to money, you know, that that's another cost that you have to think about. 00:57:29.600 --> 00:57:32.960 What about AI and specifically vibe coding? 00:57:32.970 --> 00:57:37.400 I'm not a huge fan of vibe coding as a concept, although it's super fun to just come up with 00:57:37.400 --> 00:57:40.460 a crazy idea and send the agent off and go, I wonder what it does. 00:57:40.750 --> 00:57:40.880 Right. 00:57:40.980 --> 00:57:54.960 But as a form of engineering, I'm not a huge fan of it, but I do think that we're in an emerging world where there's many people who have programmed in a sense and they have an app and then they're just completely like, well, how do I get it on the internet? 00:57:55.600 --> 00:57:57.360 You know, because they're not really developers, right? 00:57:57.760 --> 00:58:02.300 And there's like, I don't Linux, I don't DevOps, but I still need it on the internet. 00:58:02.860 --> 00:58:03.800 How do I go about that? 00:58:03.980 --> 00:58:04.080 Right? 00:58:04.390 --> 00:58:04.680 What do you think? 00:58:04.760 --> 00:58:11.460 Yeah, I mean, that's, you know, a massive cohort that's, you know, kind of emerging into the new 00:58:11.700 --> 00:58:16.780 space of builders, right? And, you know, they want to host their stuff in a place where it's live, 00:58:16.940 --> 00:58:20.820 and they don't have to, you know, deal with the extra, you know, headache of figuring out, 00:58:21.060 --> 00:58:24.920 like you mentioned, what is all this Linux stuff? They kind of just want to be like, hey, I vibe 00:58:25.000 --> 00:58:29.400 coded this Flask app, this Django app, this FastAPI app, and it's really cool. And I share with all 00:58:29.500 --> 00:58:34.360 my friends, how do I do that? Right? Where do I even start? Like, what is a server? 00:58:34.720 --> 00:58:38.060 These types of questions are really a big blocker, right? 00:58:38.240 --> 00:58:42.660 And that we do serve those people as well, quite well. 00:58:42.960 --> 00:58:45.460 Yeah, I don't want to position you initially as only serving VibeCoders. 00:58:45.460 --> 00:58:49.200 It just, it seems to me like that's a whole nother area of people who really could be 00:58:49.520 --> 00:58:51.260 served from a simplified deployment experience. 00:58:51.700 --> 00:58:54.640 But there's others, you know, I'm thinking data science, honestly. 00:58:55.160 --> 00:58:59.640 There's a ton of data scientists who have built something amazing, but they're not really 00:59:00.240 --> 00:59:02.180 DevOps software engineer side. 00:59:02.320 --> 00:59:14.820 They come from more from the science side and they've done really cool work, but it's a big step to go from I've got my notebooks running to now I have a FastAPI app to now I have it on Linux. 00:59:15.740 --> 00:59:21.700 Yeah, no, there's a ton of like beginner cohorts, I would say, that, you know, are served really well. 00:59:22.600 --> 00:59:29.060 For us, we're like mainly focused on, I would say, two main cohorts of people that we see as our ICP. 00:59:29.520 --> 00:59:34.080 One is like smaller development and startup teams, right, that have a Python stack. 00:59:34.720 --> 00:59:38.360 And, you know, they want to spend most of their time in the business logic and actually 00:59:38.640 --> 00:59:42.160 serving their customers' needs and, you know, focusing on their customer feedback rather 00:59:42.260 --> 00:59:44.700 than managing their internal hosting solutions. 00:59:45.200 --> 00:59:48.760 So they, you know, wouldn't have to, you know, hire someone to do that or spend their dev 00:59:48.900 --> 00:59:50.540 time, you know, on those tasks. 00:59:51.020 --> 00:59:56.220 And then the other cohort are, you know, agencies and consultants, kind of by the same token, 00:59:56.320 --> 01:00:00.760 but they're served a little differently by one of our core features, 01:00:01.010 --> 01:00:03.640 which we kind of haven't talked about yet, which is the managed cloud. 01:00:04.380 --> 01:00:04.820 Yeah, tell us. 01:00:05.460 --> 01:00:09.000 Let's take a step back and maybe position Femoral around. 01:00:09.480 --> 01:00:14.720 On one hand, we have infrastructure to the service, EC2, DigitalOcean, Hetzner, VPCs, whatever. 01:00:14.870 --> 01:00:18.060 And on the other, maybe you just upload some code somewhere. 01:00:18.960 --> 01:00:20.040 Where do you all live? 01:00:21.120 --> 01:00:24.420 I think I may have missed a couple words in there because of internet, 01:00:24.680 --> 01:00:41.400 But I am going to say we fall kind of in the platform as a service side of things where you want a place to host your apps and you just want to be able to throw them up there without having to deal with these servers manually. 01:00:42.060 --> 01:00:42.440 Yeah, exactly. 01:00:43.120 --> 01:00:44.260 You got the gist of it. 01:00:44.310 --> 01:00:47.440 So basically, it's sort of a continuous deployment type thing. 01:00:47.600 --> 01:00:49.660 Create a certain branch, like a production branch or whatever. 01:00:50.060 --> 01:00:50.960 Connect GitHub to it. 01:00:51.220 --> 01:00:51.840 Push that branch. 01:00:52.300 --> 01:00:52.600 Off you go. 01:00:52.670 --> 01:00:54.260 Like merge a PR and it's live, right? 01:00:54.600 --> 01:00:59.080 Yeah, pretty much. And then all of your code, without having to dockerize or containerize, 01:00:59.620 --> 01:01:04.620 because of that build pipeline we built, will get built and will then deploy it onto our managed 01:01:04.820 --> 01:01:12.000 cloud, which uses our serverless orchestrator to scale up and scale down and manage the amount of 01:01:12.940 --> 01:01:18.360 instances slash containers slash VMs that are running your code relative to the amount of 01:01:18.500 --> 01:01:23.219 traffic that's coming and the amount of resources that your code base is using at any given moment 01:01:23.220 --> 01:01:26.800 will be properly provisioned to match that traffic. 01:01:26.980 --> 01:01:29.640 So that's kind of like what we do in an end-to-end sense. 01:01:30.160 --> 01:01:33.280 Code uploaded, we build, we deploy, then we scale. 01:01:33.760 --> 01:01:35.980 So those are all the things we take care of. 01:01:36.240 --> 01:01:36.780 Yeah, awesome. 01:01:37.060 --> 01:01:38.700 I see it says now in early access. 01:01:39.260 --> 01:01:41.160 How do I get early access? 01:01:41.640 --> 01:01:44.980 Yeah, you just click that button that says get started 01:01:45.100 --> 01:01:45.720 and you sign up. 01:01:45.840 --> 01:01:49.360 Yeah, there's no, really, there's no gating here. 01:01:49.740 --> 01:01:50.280 Yeah, got it. 01:01:50.540 --> 01:01:51.820 I have talked to some folks who are like, 01:01:51.840 --> 01:01:54.840 yeah, not yet, but you can apply and hopefully, so we'll see. 01:01:55.220 --> 01:01:56.380 Yeah, we're wide open. 01:01:57.200 --> 01:02:01.600 We would love for anyone who has something they'd like to deploy to come check it out. 01:02:01.860 --> 01:02:02.100 Sweet. 01:02:02.460 --> 01:02:02.780 Startup Row. 01:02:03.200 --> 01:02:03.600 What do you think? 01:02:03.720 --> 01:02:04.160 How'd you get there? 01:02:04.300 --> 01:02:06.100 Why do you put the time and energy into that? 01:02:06.400 --> 01:02:08.140 Yeah, Startup Row, it's pretty awesome. 01:02:08.440 --> 01:02:15.360 You know, I was in contact with some of the organizers and, you know, I don't think I really 01:02:16.380 --> 01:02:20.340 had a strong sense of what Startup Row would be like before I actually got there. 01:02:21.420 --> 01:02:23.580 given that it was also my first PyCon at the same time. 01:02:23.670 --> 01:02:25.420 So there's a lot of new happening. 01:02:26.430 --> 01:02:29.560 But I would say I did it, you know, 01:02:29.630 --> 01:02:32.760 to just kind of get more involved in the community, right? 01:02:33.020 --> 01:02:35.620 Like as the company were, you know, made for Python dev. 01:02:35.720 --> 01:02:38.600 So naturally I want to be, you know, involved in the community. 01:02:39.000 --> 01:02:42.940 And then once we got there, it was pretty spectacular, you know, 01:02:43.280 --> 01:02:46.280 got to meet and talk to like so many developers 01:02:46.660 --> 01:02:47.780 and people with different problems. 01:02:48.070 --> 01:02:49.360 And, you know, people would come up and say, 01:02:49.520 --> 01:02:52.440 oh my God, this is something that I really need right now. 01:02:52.540 --> 01:02:53.740 And then you'd have people that come up and be like, 01:02:54.420 --> 01:02:55.720 this is so foreign to me. 01:02:56.040 --> 01:02:57.700 I don't even understand why I would need this. 01:02:57.860 --> 01:03:00.440 So there's like, you just get the whole spectrum 01:03:00.900 --> 01:03:02.480 and it gives you a really good perspective. 01:03:02.760 --> 01:03:03.140 That's cool. 01:03:03.240 --> 01:03:04.120 I'm glad you had a good experience. 01:03:04.400 --> 01:03:06.300 I don't go to PyCon every single year, 01:03:06.340 --> 01:03:08.740 but when I do, I always really enjoy being there. 01:03:09.220 --> 01:03:11.280 It's great to just be with people in the community 01:03:11.420 --> 01:03:13.060 doing really smart and interesting things. 01:03:13.520 --> 01:03:16.620 Yeah, no, I learned a lot more about the community 01:03:16.780 --> 01:03:18.720 than I knew before I got there. 01:03:19.060 --> 01:03:22.620 Yeah. I think it's also, as a startup, it must be interesting to talk to people and go, 01:03:22.690 --> 01:03:26.680 I don't think I need you. What is this? Why do I need this? Because either it helps you focus in 01:03:26.780 --> 01:03:31.720 on your ideal customer and say you're actually outside of that slice of the world, or position 01:03:31.830 --> 01:03:35.840 your messaging a little bit, right? Like one of the big stories of startups is talk to customers, 01:03:36.040 --> 01:03:40.660 talk to customers, like so much easier said than done though. Yeah, no, definitely. You're right 01:03:40.670 --> 01:03:47.320 about that. I would say we were able to validate our ICP because the most interest we got was from 01:03:47.340 --> 01:03:52.780 smaller startup teams and from agencies slash freelancers or consultants that were, you know, 01:03:53.120 --> 01:03:56.280 dealing with clients and had all these staging environments that they had to manage and all this 01:03:56.440 --> 01:04:01.780 extra infra. So like kind of validated that in a sense, but, you know, I guess at the same time 01:04:02.460 --> 01:04:06.740 invalidated outside of our ICP simultaneously because of those conversations I talked about 01:04:06.840 --> 01:04:11.920 earlier. I hadn't really thought about the agency side, but a lot of what happens with the agency is 01:04:12.040 --> 01:04:15.500 somebody comes in and builds something really nice and they get it working and then they hand it off 01:04:15.520 --> 01:04:18.940 to somebody who's presumably not quite as familiar with the tech 01:04:19.110 --> 01:04:20.740 because otherwise they might have just built it themselves. 01:04:21.280 --> 01:04:23.840 So the concept of, look, you just pushed to this branch. 01:04:24.220 --> 01:04:24.460 Yeah. 01:04:24.720 --> 01:04:26.360 That's got to go over well as a handoff. 01:04:26.700 --> 01:04:28.020 Right. It's great for a handoff. 01:04:28.050 --> 01:04:30.040 And it's also great for managing, you know, 01:04:30.220 --> 01:04:32.200 like mock-ups and staging environments, right? 01:04:32.460 --> 01:04:35.180 Like say you have, you know, a number of clients 01:04:35.480 --> 01:04:39.060 and you send them a mock-up or two or a live demo every week. 01:04:39.600 --> 01:04:41.900 And then you have to manage kind of this extra staging environment 01:04:42.500 --> 01:04:43.980 for every single one of these clients. 01:04:44.340 --> 01:04:49.160 it becomes a lot more infrastructure work for you as a company compared to like, say, 01:04:49.240 --> 01:04:52.140 you're just a startup and you just have one product and, you know, you need one staging 01:04:52.340 --> 01:04:56.640 environment for that one product. Right. So like for agencies, this kind of infowork can kind of 01:04:57.210 --> 01:05:01.700 increase pretty rapidly as the clients increase per client almost. Yeah. Right. Interesting. And 01:05:01.800 --> 01:05:05.660 then the scaling, right. Between each client. Right. So there's like, you know, I could get into it, 01:05:05.760 --> 01:05:11.080 but we only have so much time. Indeed. So how did you sort of told me how your experience was, 01:05:11.140 --> 01:05:14.200 but as a startup, how was startup row for you? 01:05:14.540 --> 01:05:16.400 Yeah, I mean, it was great. 01:05:16.660 --> 01:05:18.760 We, you know, we've made some really good connections 01:05:19.560 --> 01:05:22.780 for potential partners going forward, potential customers. 01:05:23.080 --> 01:05:25.280 So I would say it really helped grow the business 01:05:26.120 --> 01:05:28.080 and gave us, you know, like, I guess I already said this, 01:05:28.180 --> 01:05:31.000 but gave us perspective on like what certain, 01:05:31.560 --> 01:05:34.240 you know, personas perceive of the product, 01:05:34.480 --> 01:05:36.140 you know, given their context and their background. 01:05:36.560 --> 01:05:39.060 I found it very interesting to just walk around the expo floor 01:05:39.360 --> 01:05:43.100 and kind of just get a sense of what have people come to promote, right? 01:05:43.150 --> 01:05:46.340 You get one time a year to come talk about something and what is their, 01:05:46.760 --> 01:05:48.240 what does their banner say? 01:05:48.670 --> 01:05:49.520 Who decided to come? 01:05:49.570 --> 01:05:50.780 Who decided not to come? 01:05:50.830 --> 01:05:54.160 And yeah, it's really, it's a neat experience. 01:05:54.540 --> 01:05:54.860 Definitely. 01:05:55.600 --> 01:05:58.720 Tech, how much do you want to talk about tech behind Femoral? 01:05:58.860 --> 01:06:01.800 I think it'd be interesting, but I don't want to make you give away secrets. 01:06:02.230 --> 01:06:04.180 So tell us what you're comfortable sharing. 01:06:04.540 --> 01:06:05.280 Yeah, it's fun. 01:06:05.940 --> 01:06:17.680 So I guess I can kind of talk about the functional experience of what the tech enables, which is that, you know, we built this build pipeline. 01:06:18.340 --> 01:06:25.540 And essentially what we're doing, in a sense, is, you know, at a high level, going through your code base and figuring out like, okay, what are the tools you're using? 01:06:25.740 --> 01:06:26.660 What are the libraries you're using? 01:06:26.900 --> 01:06:27.740 What do we need to package? 01:06:27.980 --> 01:06:28.780 What do we need to build? 01:06:29.240 --> 01:06:31.340 And, you know, like that's quite straightforward, right? 01:06:31.460 --> 01:06:32.540 You're just kind of parsing code. 01:06:33.040 --> 01:06:34.240 It's largely a solved problem. 01:06:35.380 --> 01:06:42.320 But then, you know, I guess the more technically interesting side of things is the like serverless orchestration that we've written in the managed cloud. 01:06:42.840 --> 01:06:56.340 And so that's kind of based off of, you know, these like fast starting VMs that we use and some kind of, you know, way to take advantage of, you know, really fast cold starts with those fast starting VMs. 01:06:56.680 --> 01:07:02.920 Right. So like, you know, when someone does have something deployed serverlessly and it is scaled to zero, well, when they hit it, they don't want to be sitting there for a minute. 01:07:03.260 --> 01:07:06.920 right waiting for some instance to spin up right so like a really interesting cost level right 01:07:07.040 --> 01:07:11.720 because if you want to keep one running so it's not a cold start but that that that costs money 01:07:12.260 --> 01:07:16.280 but if you want to turn it off maybe it's well i don't want to wait 30 seconds for the first request 01:07:16.500 --> 01:07:21.500 right right and that's like really uh what we wanted to avoid and what generally happens with 01:07:21.620 --> 01:07:25.140 persistent compute so that's like one of the reasons that we as a company decided to build 01:07:25.280 --> 01:07:30.220 on serverless compute um in order to like you know drive this experience where you can have all these 01:07:30.240 --> 01:07:33.820 things deploy and they've all scaled to zero, but you're not, you know, waiting on a cold start when 01:07:33.940 --> 01:07:38.260 you do actually want to use something. And that's kind of powered by these fast starting VMs that 01:07:38.360 --> 01:07:43.920 we've kind of built everything on top of. Okay. What can I host there? Look at your website. It 01:07:44.000 --> 01:07:49.300 sounds like the standard Python stack sort of thing. Yeah. It's like, I mean, that's kind of 01:07:49.320 --> 01:07:55.920 like just the biggest subset of the set of things we can deploy on Ephemeral. Mostly it's anything 01:07:55.940 --> 01:08:01.460 written in python exposed with an api and that you wouldn't need gpu compute for because right now we 01:08:01.560 --> 01:08:07.400 don't have for gpu compute for many reasons that that's gpu is really interesting it's as soon as 01:08:07.400 --> 01:08:10.800 you go down that path it's like well you could start at a thousand dollars and go up from there 01:08:11.020 --> 01:08:17.080 rather than ten you know it's like yeah it's tough to offer a reasonable price point with gpus yeah 01:08:17.200 --> 01:08:23.460 i'm sure that it is um so one thing i see the web frameworks and they all look real common of 01:08:23.460 --> 01:08:26.660 Of course, if you can do all those three, then you can do others, I imagine. 01:08:27.100 --> 01:08:28.960 What's the story for data access? 01:08:30.040 --> 01:08:32.299 Databases, Postgres, SQLite, Mongo, others? 01:08:32.299 --> 01:08:35.920 Yeah, so I guess this kind of feeds into our roadmap, 01:08:36.380 --> 01:08:40.560 which is we're now working on building serverless Postgres, 01:08:40.650 --> 01:08:43.839 like managed serverless Postgres, to add right into Femoral. 01:08:44.060 --> 01:08:46.560 So you deploy your serverless web app, 01:08:46.880 --> 01:08:48.960 and right next to it, you deploy your serverless Postgres. 01:08:49.420 --> 01:08:50.160 They talk to each other. 01:08:50.480 --> 01:08:51.420 It's all in the same platform. 01:08:51.940 --> 01:08:53.319 Everything's managed under the same billing. 01:08:53.460 --> 01:08:54.460 So it's really straightforward. 01:08:54.799 --> 01:08:57.200 So yeah, right now, if you were to deploy in Femoral, 01:08:57.620 --> 01:08:59.580 your database would live outside of our compute. 01:09:00.040 --> 01:09:02.319 But we are very soon bringing that 01:09:02.589 --> 01:09:04.180 into the scope of what Femoral does. 01:09:04.299 --> 01:09:04.600 Got it. 01:09:04.730 --> 01:09:05.960 Do I get to pick my cloud? 01:09:06.009 --> 01:09:10.120 Do I say EC2 or like, sorry, AWS or Azure 01:09:10.620 --> 01:09:12.120 or DigitalOcean or whatever? 01:09:12.230 --> 01:09:15.220 Do I choose that then pick up a managed database 01:09:15.450 --> 01:09:16.020 in that location? 01:09:16.240 --> 01:09:18.580 Or how do I relate those two things? 01:09:18.960 --> 01:09:22.620 Yeah, so you can choose to locate your database. 01:09:22.779 --> 01:09:28.759 like when our database managed progress is offered, you can choose to locate that in whichever region, 01:09:29.799 --> 01:09:34.420 basically any AWS region. Although there are obviously other hyperscalers that we could offer 01:09:34.450 --> 01:09:41.080 that in right now. And in terms of the actual web app compute itself, that is not necessarily 01:09:42.339 --> 01:09:50.719 provisioned into a specific region. But we are working on how we might communicate that to the 01:09:50.620 --> 01:09:55.700 user. Sounds good. And I guess anything else on the roadmap that you want to touch on, or is that 01:09:55.890 --> 01:10:00.080 pretty much covered the database? Yeah. I mean, basically, right. You have the stateless compute, 01:10:00.250 --> 01:10:03.900 and then obviously you need some states somewhere, right? So that's all we're working on, right, 01:10:03.900 --> 01:10:10.260 is the Postgres. People want, obviously, Redis, some kind of key value. So managed Redis will 01:10:10.760 --> 01:10:16.920 probably come with Postgres. Yeah. Shinmaya, congrats on starting Femoral and best of luck 01:10:16.940 --> 01:10:20.560 to you guys and nice to meet you at PyCon. Thanks for coming on the show. Yeah, for sure. Nice 01:10:20.620 --> 01:10:27.580 meeting you too. See you later. Now we meet Capiccio in Beyond, its co-founder. Hey, Beyond, 01:10:27.760 --> 01:10:31.400 welcome to Talk Python To Me. Yeah, thank you so much, Michael. Really great to be on the show. 01:10:31.740 --> 01:10:37.160 Yeah, it's fabulous to have you be part of our startup row segment, stroll down startup lane, 01:10:37.380 --> 01:10:43.280 if you will. I think it's such a neat idea that PyCon, the PyCon organizers, Jason and Shea and 01:10:43.300 --> 01:10:48.340 everyone else put that together. And so it's just a neat look into what's up and coming in the Python 01:10:48.680 --> 01:10:55.180 space. So we're going to talk about Capicio. Capicio. Yeah. Capicio. I love it. This is your 01:10:55.540 --> 01:11:00.640 business. And we're going to dive into why you want Startup Row and all those things. But before we do, 01:11:00.860 --> 01:11:05.599 quick introduction. Who are you? Yeah, sure. Thank you. So my name is Beyond Denote. I am the founder 01:11:05.620 --> 01:11:14.120 of Capicio, we refer to ourselves as the authority layer for AI agents. So identity policy, basically 01:11:14.390 --> 01:11:20.340 to be able to secure agentic environments so that they're ready for production and your 01:11:20.500 --> 01:11:25.160 SecOps teams aren't freaking out when you're trying to deploy agents to prod. 01:11:25.340 --> 01:11:30.719 It's really tricky, right? We've got this non-deterministic thing running, but it's supposed 01:11:30.660 --> 01:11:35.660 to be operating in a professional or legally structured environment, right? 01:11:36.040 --> 01:11:36.120 Yeah. 01:11:36.380 --> 01:11:36.480 Yeah. 01:11:36.480 --> 01:11:37.780 It's a massive challenge. 01:11:38.560 --> 01:11:43.800 It's blocking a lot of POCs right in its tracks at the moment, or even some organizations 01:11:43.980 --> 01:11:45.880 from not even considering AI solutions. 01:11:46.720 --> 01:11:49.520 So we're trying to bridge that gap and solve that problem. 01:11:49.800 --> 01:11:54.980 You know, I'm reminded of some stat that was probably just made up or whatever, or very 01:11:55.220 --> 01:11:58.300 vague, like 97% of all AI projects fail. 01:11:59.360 --> 01:12:02.000 And I have no idea what that even means. 01:12:02.220 --> 01:12:06.620 Does that mean like they tried to add AI to an email client and they just, it couldn't 01:12:06.640 --> 01:12:07.040 be added? 01:12:07.240 --> 01:12:11.160 Like, no, my email client just has a very bad representation of AI in it. 01:12:11.560 --> 01:12:13.080 Yeah, I have no idea. 01:12:13.580 --> 01:12:14.280 Yeah, no, no. 01:12:14.500 --> 01:12:15.900 I've seen that stat as well. 01:12:16.060 --> 01:12:21.880 I think the source was some MIT paper that ended up maybe not being quite as well researched. 01:12:22.000 --> 01:12:23.420 So I don't know what the real stat is. 01:12:23.600 --> 01:12:25.940 Maybe we're somewhere close, but yeah. 01:12:26.140 --> 01:12:28.620 Yeah, I think it's a much higher level of success. 01:12:28.900 --> 01:12:33.100 but things like what you all are building are sort of key to that success, right? 01:12:33.400 --> 01:12:38.900 Yeah, absolutely. It's a whole new ecosystem. I mean, if we go back to the early 2000s, 01:12:38.910 --> 01:12:44.180 we invented this thing called the internet and we had all these great ideas, but there was a lot 01:12:44.180 --> 01:12:49.040 of infrastructure still missing to be able to realize that. And at the time we didn't really 01:12:49.300 --> 01:12:55.840 know what infrastructure we really needed yet. With the whole agentic era, I think we've hit a reset 01:12:55.860 --> 01:13:01.220 button on a lot of that except we're a lot more technologically mature so i think we teams know 01:13:01.380 --> 01:13:06.060 what they need and what they're missing and we're in this build race to to be able to close those 01:13:06.280 --> 01:13:11.800 gaps interesting i do think it's just absolutely the wild wild west is not the right term but it's 01:13:11.820 --> 01:13:16.660 just an unknown we don't know what we need we don't even necessarily to some degree know what 01:13:16.720 --> 01:13:22.180 we don't know for some of these products but it's changing fast i was sort of thinking about that 01:13:22.100 --> 01:13:26.660 that one when i threw out when your response to that that stat i threw out is like how much of 01:13:26.660 --> 01:13:33.120 that is based on 2024 ai you know what i mean right it moves so fast i mean even six months ago and 01:13:33.180 --> 01:13:38.660 you're looking at a different beast you know it's uh it's incredible the iteration pace is so fast 01:13:39.080 --> 01:13:45.400 yeah so why'd you start capuchio yes uh good question so um i obviously had been following 01:13:45.450 --> 01:13:51.100 the ai space um you know jumping on everything that that i could to be able to learn more and get 01:13:51.080 --> 01:13:51.660 into the space. 01:13:52.620 --> 01:13:54.340 But I wanted to do something meaningful. 01:13:54.420 --> 01:13:56.020 I wanted something infrastructure level. 01:13:56.460 --> 01:13:59.940 And then when Google released the first iteration 01:14:00.240 --> 01:14:03.120 of the A2A protocol, the agent to agent protocol, 01:14:03.380 --> 01:14:06.420 I think it was somewhere around March last year, March 2025, 01:14:08.780 --> 01:14:10.240 a few light bulbs started going off. 01:14:10.340 --> 01:14:14.320 Because here we were creating a new standard for agents 01:14:14.740 --> 01:14:18.940 not to communicate with MCP servers or human agent. 01:14:19.160 --> 01:14:23.320 This was agent to agent communication, which we knew 01:14:23.440 --> 01:14:26.520 was coming, but the concept blew my mind. 01:14:27.270 --> 01:14:30.400 And then looking at DeepMind, Google DeepMind also 01:14:30.540 --> 01:14:33.340 released a few papers on intelligent delegation, 01:14:34.540 --> 01:14:35.300 agent to agent delegation. 01:14:35.980 --> 01:14:39.060 So looking at that, from where we were, 01:14:39.180 --> 01:14:41.960 we're just using basically LLMs and chatbots 01:14:42.240 --> 01:14:45.720 to a layer where we had agents self-orchestrating 01:14:46.690 --> 01:14:47.860 and trying to do that at scale. 01:14:48.620 --> 01:14:51.160 That's where, like I explained, early 2000s internet, 01:14:51.500 --> 01:14:52.640 we're missing a lot of infrastructure. 01:14:53.600 --> 01:14:55.900 And for now, being able to see that realization 01:14:56.160 --> 01:14:58.400 for that vision, there's a lot of infrastructure missing. 01:14:58.740 --> 01:15:00.980 So that's where I started building, 01:15:01.800 --> 01:15:03.660 particularly around the trust layer, 01:15:04.460 --> 01:15:05.340 the authority layer, 01:15:06.120 --> 01:15:08.840 trying to make sure that agents can communicate 01:15:09.400 --> 01:15:10.980 securely, safely together, 01:15:11.220 --> 01:15:12.740 and that we can facilitate that 01:15:12.920 --> 01:15:14.840 with all the missing layers that we have right now. 01:15:15.000 --> 01:15:17.440 Well, it's definitely a challenging problem, right? 01:15:17.820 --> 01:15:23.140 We used to write tests or have security scanners, but now, what are some of the new challenges, 01:15:23.340 --> 01:15:28.040 I guess, that you see out there that are different than prior stages? 01:15:28.620 --> 01:15:30.100 Yeah, good question. 01:15:30.420 --> 01:15:35.180 So one of the challenges has to do with identity. 01:15:35.860 --> 01:15:40.140 When I started working on the problem, I was focused more on setting guardrails for AI 01:15:40.480 --> 01:15:45.380 agents, helping them to communicate together, being able to learn more about each other, 01:15:46.440 --> 01:15:51.600 But being able to grant every agent an identity was very important. 01:15:52.540 --> 01:16:04.320 Right now, it seems like the current trend is that we're trying to expand the known models around IAM into agentic identity, which is sort of a band-aid right now. 01:16:04.320 --> 01:16:09.800 But one of the biggest restrictions is that those are usually within a singular domain. 01:16:10.320 --> 01:16:22.400 And cross-domain open web communication, being able to extend that identity concept to really the future that we have with agentic AI on the open web, that becomes a big challenge. 01:16:23.360 --> 01:16:30.820 And then along with that, of course, we have what we're referring to as the trust layer. 01:16:31.800 --> 01:16:34.940 Just like in human society, we have people we interact with. 01:16:35.520 --> 01:16:44.360 And the first thing we even do before we buy a new pair of shoes or a restaurant is we check reviews, you know, on the merchant, on the actual product. 01:16:45.300 --> 01:16:49.340 And that mechanism doesn't really exist for agents right now. 01:16:49.560 --> 01:16:58.040 So in a way, there's some of those primitives that are missing to be able to really scale that Agentech future up to full speed. 01:16:58.400 --> 01:16:59.180 Going to be weird, right? 01:16:59.400 --> 01:17:00.100 Yeah, absolutely. 01:17:00.460 --> 01:17:06.400 All these things, you'll have to just drop in and check on your team of agents and see how things are going. 01:17:06.880 --> 01:17:07.200 You know what? 01:17:07.650 --> 01:17:15.360 I can easily see a product manager or some kind of manager agent that you go and work with. 01:17:15.640 --> 01:17:18.640 And then it's like, look, how's the whole thing going? 01:17:18.840 --> 01:17:19.680 How's the team looking? 01:17:20.290 --> 01:17:22.400 And then you scope it out. 01:17:22.500 --> 01:17:23.880 You all go work on this. 01:17:23.880 --> 01:17:24.520 Get back to me later. 01:17:24.620 --> 01:17:25.680 It's going to be weird. 01:17:26.280 --> 01:17:28.680 Yeah, we're definitely heading to that future. 01:17:28.760 --> 01:17:40.800 I mean, we've maybe seen it in movies before, and it seemed like so far off, you know, but we're all definitely going to have our personal agent that just kind of knows what's going on. 01:17:41.160 --> 01:17:45.140 And I don't think it's as far off as what a lot of us are imagining. 01:17:45.840 --> 01:17:48.060 No, I don't either. It's going to be odd. 01:17:48.320 --> 01:17:50.440 Okay, so Startup Pro, that's where we met. 01:17:51.120 --> 01:17:54.600 And while you all were at PyCon, why apply to Startup Pro? 01:17:55.020 --> 01:17:55.480 How'd you get there? 01:17:55.760 --> 01:17:56.640 Yeah, no, thanks. 01:17:56.720 --> 01:17:57.140 Good question. 01:17:58.340 --> 01:18:16.700 So I have to be honest here because like a lot of folks who are now using Python and fluent in Python, I sort of backed into the Python community because around the rise of AI, obviously Python, I mean, we can refer to it as the native language for AI pretty much, right? 01:18:16.800 --> 01:18:21.100 So started using a lot of Python each and every day. 01:18:22.100 --> 01:18:24.800 And then this product that we built obviously has the SDK. 01:18:24.960 --> 01:18:29.680 It has an MCP plugin for MCP server harnesses as well. 01:18:31.180 --> 01:18:33.500 And then one day I got pinged by Jason. 01:18:34.600 --> 01:18:36.720 You know, Jason and Shay organizing Startup Row. 01:18:37.420 --> 01:18:41.400 And he gets pinged and he's like, hey, man, have you heard about Startup Row on PyCon? 01:18:41.560 --> 01:18:42.840 I think you should really apply. 01:18:43.400 --> 01:18:44.880 And I was like, wow, okay. 01:18:45.980 --> 01:18:46.360 Checked it out. 01:18:46.660 --> 01:18:52.040 learn about PyCon I'd never been before and Startup Row and it just seemed like a great 01:18:52.270 --> 01:18:56.840 opportunity so I put the application in but like a lot of things you put in the application like 01:18:57.020 --> 01:19:03.220 whatever you know let's see what comes of it and yeah it was a little while later got pinged to say 01:19:03.440 --> 01:19:10.280 congratulations so that's how we landed on Startup Row at PyCon and man what a what a gift 01:19:10.420 --> 01:19:15.140 what a privilege it really was a it was an awesome experience. Was that your first PyCon attendance? 01:19:15.340 --> 01:19:33.180 It was. It was my first PyCon. I've been to a lot of other conferences. Even my partner who was with me at the event, he was just saying how he's worked so many other conferences and the vibe was just so different. 01:19:33.340 --> 01:19:36.760 You know, when you're talking to people at PyCon, everybody was chill. 01:19:37.120 --> 01:19:38.160 You were talking builder. 01:19:38.540 --> 01:19:40.120 It was builder to builder conversations. 01:19:40.640 --> 01:19:42.380 Hey, you know, what are you working on? 01:19:42.980 --> 01:19:44.280 How are you solving this problem? 01:19:45.200 --> 01:19:47.780 And there were just a lot of real conversations. 01:19:48.240 --> 01:19:49.140 The vibe was great. 01:19:49.440 --> 01:19:50.340 I really enjoyed it. 01:19:50.500 --> 01:19:53.720 Yeah, I've been to a lot of different technology conferences as well. 01:19:53.860 --> 01:19:54.960 And it is certainly unique. 01:19:55.480 --> 01:19:56.640 And I mean that in a good way. 01:19:56.880 --> 01:19:57.980 Yeah, yeah, no, absolutely. 01:19:58.460 --> 01:20:02.060 So one of the things I think is really critical for startups is getting feedback, 01:20:02.560 --> 01:20:07.400 talking to potential or active customers. And the advice is, well, you really got to talk to 01:20:07.660 --> 01:20:12.580 customers. It's so hard to do. But on Startup Pro, you get to kind of iterate on that over and over. 01:20:12.900 --> 01:20:17.100 Did you get some good feedback? Yeah, absolutely. We weren't sure how many... 01:20:18.320 --> 01:20:22.760 This was sort of a big validation test for us because we were curious about the timing of the 01:20:22.960 --> 01:20:28.440 market. Is the market ready to have this conversation? And it was interesting from 01:20:28.460 --> 01:20:33.440 developers point of view most to be honest were like oh yeah we got to do something about this we 01:20:33.580 --> 01:20:38.640 haven't kind of got there yet um but then from from the folks who were working inside larger 01:20:38.880 --> 01:20:43.400 enterprises or the ones who have very advanced solutions they were already at the point where 01:20:43.520 --> 01:20:49.080 they they totally got it like wow okay this is awesome we need to check this out um so i mean 01:20:49.160 --> 01:20:53.260 we're following up now we got a quite a few design partner opportunities which is the point that we're 01:20:53.320 --> 01:20:56.520 we're at right now we're looking to work with teams who are experiencing this problem 01:20:57.240 --> 01:21:03.120 And, you know, where they can start running our product and find where we need to extend it to make it work for them. 01:21:03.320 --> 01:21:04.080 What about the tech? 01:21:04.800 --> 01:21:09.400 Are you giving us a peek inside what core technologies you all are using? 01:21:09.740 --> 01:21:10.240 Yeah, sure. 01:21:10.420 --> 01:21:10.680 Absolutely. 01:21:11.920 --> 01:21:20.060 So because it's an infrastructure project and there's cryptographic verification of identities, a lot of handshakes like that going on. 01:21:20.560 --> 01:21:24.220 We've actually built the core of it in Go. 01:21:24.400 --> 01:21:25.960 It's a Golang, you know, library. 01:21:26.340 --> 01:21:35.780 And then we have all the wrappers and all the stuff that would benefit from native latency in whatever target SDK language. 01:21:36.080 --> 01:21:39.640 So Python being the main one, that's the only one we have right now. 01:21:42.100 --> 01:21:47.760 So basically, what ends up happening, the flow is you can use our SDK, one line of code. 01:21:47.820 --> 01:21:51.780 We literally, that's why we had it on our banner at our booth. 01:21:52.140 --> 01:21:57.540 one line of code to be able to connect, register an identity for that agent for an MCP server. 01:21:58.460 --> 01:22:03.520 That's then cryptographically verified because the private key is stored right on the agent, 01:22:03.720 --> 01:22:09.200 right on the server. And then with that identifier, you can now be able to create policy 01:22:09.900 --> 01:22:16.340 around the tools and around the agent. What tools is it allowed to invoke? What are blocked? 01:22:17.020 --> 01:22:20.600 what group policy, what organizational policy belongs 01:22:20.980 --> 01:22:23.360 to this agent or server. 01:22:23.900 --> 01:22:25.440 And the neat part in how we designed it 01:22:25.680 --> 01:22:29.620 is those are all policies that are compiled into OPA bundles 01:22:29.920 --> 01:22:33.000 and are actually cached right to the agent or right to the server. 01:22:33.260 --> 01:22:37.140 So to be able to respond to requests is super fast. 01:22:37.280 --> 01:22:41.600 It's a sub 10 millisecond because usually when we use security, 01:22:41.660 --> 01:22:43.240 we think, oh, well, there goes my latency. 01:22:43.960 --> 01:22:45.400 So that was a problem we wanted to solve. 01:22:45.640 --> 01:22:53.460 But that in essence is as far as we've gotten in our roadmap right now, we are working on a full public RFC stack. 01:22:54.000 --> 01:23:00.200 So our next two RFCs, we're working on the intent layer, which is a whole other fun conversation. 01:23:00.350 --> 01:23:02.980 But we're determined to try and crack that as well. 01:23:03.100 --> 01:23:03.700 Yeah, yeah. Very cool. 01:23:04.010 --> 01:23:08.960 You're not the only startup on Startup Row building the core server side components with Go. 01:23:09.320 --> 01:23:11.440 Yes. And that's what we found out. 01:23:11.640 --> 01:23:18.320 You know, when we were coming to PyCon, thinking Python community, we were like, okay, don't talk about Go. 01:23:18.540 --> 01:23:19.040 This is Python. 01:23:20.440 --> 01:23:21.140 Don't say anything. 01:23:22.440 --> 01:23:26.400 And then we got some questions like you just asked, you know, were they really pressed? 01:23:26.760 --> 01:23:28.680 And they were like, okay, no, that makes sense. 01:23:28.730 --> 01:23:29.160 We get it. 01:23:29.300 --> 01:23:33.740 That's like, you're like this product or that product, that product that uses a similar pattern, you know? 01:23:33.880 --> 01:23:37.560 So, yeah, it wasn't an unusual way to implement it. 01:23:37.780 --> 01:23:43.120 Python has a really interesting performance story and relationship with other native build 01:23:43.340 --> 01:23:44.980 technologies or native languages. 01:23:45.480 --> 01:23:51.020 Like, for example, Python itself runs on CPython, which is primarily implemented in C, not Python. 01:23:51.500 --> 01:23:54.720 A lot of the ways that Python is made fast these days is through Rust. 01:23:55.360 --> 01:23:59.600 So when you end up at a performance critical section, it's like there's a layer of Python 01:23:59.740 --> 01:24:00.760 and then it's Rust, right? 01:24:00.960 --> 01:24:06.140 Like my web server, the application server for my website and my course website, all the 01:24:06.220 --> 01:24:06.360 things. 01:24:06.880 --> 01:24:11.800 It's built in Rust and then it just delegates out to Python, out to Flask, basically. 01:24:12.080 --> 01:24:17.140 So I think Python people are more than other technologies used to. 01:24:17.440 --> 01:24:23.460 So there's this core of something and then this sort of mix of choose the right thing at the right place. 01:24:23.680 --> 01:24:25.080 Yeah, yeah, no, absolutely. 01:24:25.360 --> 01:24:28.200 And Rust was a consideration for me. 01:24:28.380 --> 01:24:32.240 I came from a C background and Go just really intrigued me. 01:24:32.500 --> 01:24:36.020 So we ended up building it and Go in and it's been a good result. 01:24:37.239 --> 01:24:53.180 But yeah, I mean, especially when you're doing things like cryptographic verification and you're doing a lot of stuff over again, you wanting to scale to support many different languages, it does make sense to choose a core and then build around that. 01:24:53.260 --> 01:25:16.180 But we're constantly tweaking, trying to make it still a really native feel for whatever target platforming, including Python, so that you can use all your customary tools, have a similar experience in your IDE, and now be able to have a way also that whatever agent you're building with, that they understand the structure and the semantics as well. 01:25:16.740 --> 01:25:18.480 One more question before I let you go. 01:25:18.740 --> 01:25:21.300 Oh, I imagine there's also a lot of Docker in there, right? 01:25:21.880 --> 01:25:22.560 Yeah, absolutely. 01:25:22.660 --> 01:25:26.140 So that was part of our stack as well. 01:25:26.680 --> 01:25:37.920 One of the main reasons we introduced Docker was because a lot of enterprises, when setting up a tool like this, they'll test it on your cloud instance and be like, oh, yeah, that's great. 01:25:38.160 --> 01:25:39.520 But can we put this in our private cloud? 01:25:40.660 --> 01:25:41.660 Can we make it air-gapped? 01:25:41.880 --> 01:25:47.200 I mean, I'm hearing a lot of stories about air-gapped AI solutions and whether they actually have a future or not. 01:25:47.900 --> 01:25:55.060 But either way, we implemented around Docker so that it's an easy pull and set up to be able to run it. 01:25:55.220 --> 01:26:01.780 I do think that there's, it maybe won't be the most common way, but I think there's probably some juice in the air-gapped AI. 01:26:02.140 --> 01:26:05.040 Because with the air-gapped AI, you're going to need your own local models. 01:26:05.440 --> 01:26:17.480 But, you know, for example, a PyCon and an NVIDIA plus Anaconda combo, they had their little AI cubes, you know, their sparks doing a really cool live AI demo. 01:26:17.680 --> 01:26:21.020 And if you were a big organization, you needed, you really needed privacy. 01:26:21.020 --> 01:26:27.260 I could see buying a 10 GPU cluster and just setting that up and make it go locally. 01:26:27.740 --> 01:26:34.320 Yeah, it was interesting how many devs we spoke to who are working in highly sensitive environments. 01:26:34.720 --> 01:26:43.260 There was even one developer who works with the voice and data control recorders for airplanes. 01:26:44.660 --> 01:26:49.240 Obviously, situations like that, they very hesitantly introduce AI. 01:26:49.330 --> 01:26:53.580 And if they do, it has to absolutely be very air gapped and sandboxed. 01:26:54.100 --> 01:27:04.360 So, yeah, like you said, there's some applications where that is still a requirement, even though for most of us who have been on the cloud for however long feel it's a bygone era, but it really is not. 01:27:04.420 --> 01:27:21.000 Yeah. And if I do, I'm not a believer of, oh, the AI bubble is going to pop any day now. But I do think if the pricing changes, all of a sudden, maybe buying a $5,000 machine is the economical choice in the long term. So in which case, that also comes back to that. 01:27:21.060 --> 01:27:45.920 Yeah, yeah. And that's an interesting concept too, because obviously what we're paying today, even though we may bulk at it, it is heavily subsidized. We all know that. So either the subsidies run out and hopefully compute has a lot cheaper so that we can keep running or, you know, we have to make some tough choices about what we want to invest in and what projects are worth it. Or like you say, investing in our own hardware to run our own local LLMs. 01:27:46.040 --> 01:28:00.740 Yeah, we don't have time to go super down this, but I do think that this extreme load on compute and kind of the subsidizing as well is a forcing function for creating much more efficient compute and forcing the models to execute more efficiently. 01:28:01.200 --> 01:28:08.680 And I can certainly see a way where we just innovate our way through to where the actual price falls back to close to what we're paying. 01:28:09.260 --> 01:28:17.080 As a data point, the latest NVIDIA H200, I don't know what it's called, H200 or whatever, is 10 times more efficient at difference than the one before. 01:28:17.720 --> 01:28:18.820 By what, right? 01:28:20.060 --> 01:28:22.660 So you do that a few times, all of a sudden, there's not a bubble. 01:28:23.100 --> 01:28:23.660 It's back to normal. 01:28:24.240 --> 01:28:24.540 Absolutely. 01:28:24.880 --> 01:28:27.640 I mean, we've seen it before in many iterations, right? 01:28:27.840 --> 01:28:29.940 Like the need facilitates innovation. 01:28:30.200 --> 01:28:31.820 And that's probably where we're going to be now. 01:28:31.920 --> 01:28:40.860 And we're facing, I mean, the concept of world models and other stuff happening to quantum computing, which I don't know what everybody says, five years, 10 years, whoever. 01:28:42.140 --> 01:28:44.260 But I mean, all that is going to change the game. 01:28:44.500 --> 01:28:46.280 So we'll see how it all comes together. 01:28:46.660 --> 01:28:47.780 Beyond, thanks for sharing your thoughts. 01:28:48.280 --> 01:28:49.280 Congrats on Capitio. 01:28:49.840 --> 01:28:50.140 Thank you. 01:28:50.480 --> 01:28:51.020 Appreciate that, Michael. 01:28:51.400 --> 01:28:52.160 Yeah, thank you so much. 01:28:52.280 --> 01:28:52.560 Bye-bye. 01:28:52.860 --> 01:28:53.160 Take care. 01:28:54.440 --> 01:28:57.340 Our final startup is Pixel Table, started by Marcel. 01:28:58.960 --> 01:29:00.360 Marcel, welcome to Talk Python To Me. 01:29:00.780 --> 01:29:01.240 Nice to see you. 01:29:01.420 --> 01:29:02.780 Michael, thank you for having me. 01:29:03.220 --> 01:29:04.640 Great to have you on the show. 01:29:05.140 --> 01:29:09.760 Very cool that you all were part of Startup Row, and I'm excited to hear about Pixel Table, 01:29:10.160 --> 01:29:12.940 your new company that you're getting going, which is exciting. 01:29:13.440 --> 01:29:17.020 Before we dive into all that, though, you've got quite a history in the Python space. 01:29:17.360 --> 01:29:19.680 Tell people who you are and what you've been up to. 01:29:20.000 --> 01:29:21.180 Yeah, happy to. 01:29:21.660 --> 01:29:24.260 So my background is really in database internals. 01:29:24.400 --> 01:29:26.700 I did that at grad school a long time ago. 01:29:27.980 --> 01:29:31.780 I worked at a bunch of database startups in the early 2000s. 01:29:31.780 --> 01:29:37.040 I was at Google from '03 to '10, also working on scalable data infrastructure. 01:29:38.060 --> 01:29:45.120 And toward the tail end again on a new internal database system, what we called hybrid transactional 01:29:45.420 --> 01:29:47.580 analytic processing called F1. 01:29:47.920 --> 01:29:54.080 And after leaving Google, I joined Cloudera and started a project there called Apache Impala, 01:29:55.000 --> 01:29:58.180 or later it became an Apache project, a scale-out SQL engine. 01:29:58.800 --> 01:30:03.600 And as part of that, I also co-created the Parquet file format 01:30:04.090 --> 01:30:07.320 because there was no good open source columnar format at the time. 01:30:07.600 --> 01:30:11.840 So this was basically the motivation for creating Parquet 01:30:12.510 --> 01:30:16.900 as basically an open source implementation of Column.io from Google's Dremel. 01:30:17.600 --> 01:30:19.840 And so, yeah, so that's really my background. 01:30:19.880 --> 01:30:30.680 And I got interested in the, I want to say the ML space and computer vision by coincidence when I was an EIR at a venture firm and met a computer vision person there. 01:30:30.770 --> 01:30:32.540 So this was early 22. 01:30:33.360 --> 01:30:39.840 And I ended up then talking to a whole bunch of computer vision engineers and managers of engineering teams. 01:30:40.600 --> 01:30:42.660 And they're obviously all used Python. 01:30:42.970 --> 01:30:45.060 And all of that happens in the Python ecosystem. 01:30:45.720 --> 01:30:48.780 And that was kind of the motivation for creating Pixel Table. 01:30:49.280 --> 01:30:58.880 back then as a, I want to say, a unified view of data and storage and also the transformations 01:30:59.880 --> 01:31:05.280 that you need when you're doing computer vision work, in particular model data set curation, 01:31:06.260 --> 01:31:13.760 training, and so forth. So that's kind of the origin story of Pixel Table. And I'm happy to 01:31:13.970 --> 01:31:17.460 talk about what it is and how it has evolved since then. 01:31:17.820 --> 01:31:21.540 Yeah, that's very cool about all the database work that you've done in Parquet. 01:31:21.730 --> 01:31:24.800 And it's really clear how that's led you to Pixel Table. 01:31:25.360 --> 01:31:34.280 Yeah, it's I think, you know, people are still saving a bunch of CSVs and having huge disk usage because of all that and slow parsing. 01:31:34.330 --> 01:31:35.640 And they should look at Parquet, right? 01:31:36.040 --> 01:31:44.760 Yeah, I mean, Parquet has, you know, obviously since then become a real industry standard and is also the core of, you know, other industry standards such as Iceberg, etc. 01:31:45.080 --> 01:31:49.800 So it's gratifying to see that that had an industry impact. 01:31:50.040 --> 01:31:51.040 Yeah, I'm sure that it was. 01:31:51.400 --> 01:31:51.620 All right. 01:31:51.850 --> 01:31:57.600 So you talked about Pixel Table sort of being inspired by both database work and talking to all these vision folks. 01:31:58.080 --> 01:31:58.880 What exactly is it? 01:31:59.020 --> 01:32:01.020 It's like a database for images? 01:32:01.560 --> 01:32:02.600 Well, it's not. 01:32:02.800 --> 01:32:03.960 The name is a little misleading. 01:32:04.960 --> 01:32:08.460 It is, at its core, it's an OLTP database system. 01:32:08.680 --> 01:32:12.380 So think Postgres, basically something that you can use to power a web app. 01:32:12.390 --> 01:32:14.600 You can do single row inserts and so forth. 01:32:15.020 --> 01:32:15.760 It is multi-user. 01:32:16.200 --> 01:32:17.120 It is transactional. 01:32:17.940 --> 01:32:25.660 But now it's specifically meant for multimodal AI or multimodal applications and AI applications in particular. 01:32:26.080 --> 01:32:30.600 And so it has additional column types that represent multimodal data. 01:32:30.680 --> 01:32:36.080 So you have column types, document, video, audio, image, and array. 01:32:36.660 --> 01:32:37.220 Oh, that's cool. 01:32:37.520 --> 01:32:41.660 Basically, you can now create a table and put videos in there. 01:32:42.140 --> 01:32:44.860 And you would put the videos in there as URLs. 01:32:45.620 --> 01:32:52.140 And then Pixel Table would know when you need to do something with a video to actually transparently download it and so forth. 01:32:52.180 --> 01:32:53.180 So that's one thing. 01:32:54.220 --> 01:33:00.560 And then another specific aspect is that you can now also have computed columns in your table. 01:33:00.980 --> 01:33:09.220 And so in the AI world, when you're working with multimodal data, a lot of the work is really transformations and kind of data plumbing. 01:33:09.860 --> 01:33:13.700 And so this is kind of what you now can express inside the data model. 01:33:14.340 --> 01:33:23.240 And the pixel table runtime will then basically, you can basically express a complex workflow as a number of computed columns. 01:33:23.440 --> 01:33:25.080 It's basically a graph, a computational graph. 01:33:25.740 --> 01:33:28.380 And it could be things like, let's say, take a video. 01:33:28.520 --> 01:33:29.400 You have a table with videos. 01:33:29.960 --> 01:33:31.280 You want to get transcripts. 01:33:32.160 --> 01:33:33.420 First, you need to extract the audio. 01:33:33.560 --> 01:33:37.960 So the audio extraction would be a computed column with the output type audio. 01:33:38.580 --> 01:33:45.100 And it uses a UDF called Extract Audio, which uses FFmpeg or libAV. 01:33:45.430 --> 01:33:50.620 So we're using standard Python ecosystem functionality for processing this data. 01:33:51.360 --> 01:33:56.120 But you can now string it together without having to think about all the data plumbing. 01:33:56.520 --> 01:33:58.260 And the whole thing is still transactional. 01:33:58.410 --> 01:34:00.600 So you extract the audio as a computed column. 01:34:01.260 --> 01:34:06.900 And then you could invoke a transcription model to get the transcript out as an example, which would be another computed column. 01:34:07.240 --> 01:34:09.980 So you can build up these very complex graphs. 01:34:10.260 --> 01:34:16.500 And then you must have some kind of backend that holds the audio and holds the transcript and then links to it or kind of like you talked about in the video. 01:34:16.600 --> 01:34:21.940 So what we do is all basically media data is external and file based. 01:34:22.120 --> 01:34:27.500 So if you show up with, you know, let's say a petabyte of videos, we're not expecting you to upload that. 01:34:27.540 --> 01:34:28.640 That would be extremely inefficient. 01:34:29.260 --> 01:34:30.580 So we're simply referencing it. 01:34:31.620 --> 01:34:32.960 The same is true for audio files. 01:34:33.180 --> 01:34:34.500 The same is true for computed columns. 01:34:34.540 --> 01:34:43.020 You can actually even put a destination attribute on a computed column and tell it to put the media data into a bucket, as an example. 01:34:43.500 --> 01:34:54.360 But then we put the structured data into Postgres, and we also use PG vector, but we have our own transaction system and our own type system on top of it. 01:34:54.640 --> 01:35:04.360 And so when you do an insert into a pixel table table, it figures out the complete plan, like what goes into Postgres, what needs to happen in Python directly. 01:35:04.860 --> 01:35:16.160 We have a runtime system and an execution system that is now able to take this computational graph and optimize it and do parallelization and asynchronous execution and stuff like that. 01:35:16.320 --> 01:35:39.620 This is incredibly interesting. The more you're telling me about it, I think this is a really clever idea. The way it kind of turns the database into this workflow engine that just hides all the messiness of, oh, I just need the audio of this thing. So we got the audio column based on, you know, like computer column. And that's, I can see that just saving so much work, so much queue, asynchronous programming and all kinds of things. 01:35:40.060 --> 01:36:00.180 Yeah, exactly. And that's really the motivation and the application area, which is basically multimodal data processing, right? Including very complex systems that are fairly easy to now model as basically a sequence of tables and views. So, you know, for a podcaster like you, this might actually be very advantageous. 01:36:00.320 --> 01:36:17.240 Yeah, absolutely. I do transcripts and get the audio from videos and all sorts of stuff all the time. It certainly connects with me. So two things I want to talk about on your offering here, kind of the business model and just some features before we move on to startup row. One, let's go with a smaller one first. 01:36:17.360 --> 01:36:25.280 I see that you have skills for working with Pixel Table and that you can install the skills with npm or other ways. 01:36:26.020 --> 01:36:29.460 You also have startup with uv, like a startup template type thing. 01:36:30.300 --> 01:36:41.660 I think there's some really interesting trends for people, including skills and other AI enablers or accelerators in their documentation or in their projects. 01:36:42.100 --> 01:36:43.140 That's really neat. 01:36:43.320 --> 01:36:47.940 the fact that you can just get the skill so your agent can just jump right in. What's your thinking 01:36:48.100 --> 01:36:53.040 there? Yeah, exactly. I mean, this is sort of the way people are writing applications today, 01:36:53.760 --> 01:36:59.840 right? You kind of expect to use AI to generate most of the code. And so, you know, Pixel Table 01:36:59.870 --> 01:37:05.640 comes with a skill. And we actually think that Pixel Table is very agent-friendly and very AI 01:37:05.900 --> 01:37:12.039 coding-friendly because it gives you a semantic model that basically gives you type safety and 01:37:12.060 --> 01:37:19.080 also type safety of your data at rest. So it gives you a lot of guardrails and it allows you to 01:37:19.380 --> 01:37:25.460 express these complex workflows with relatively little code. So there is less room for the AI 01:37:26.070 --> 01:37:30.340 to drive the thing off a cliff, right? You're not going to, even with a very complex pipeline or a 01:37:30.340 --> 01:37:36.480 very complex workflow, it's typically no more than a few hundred lines of code because all of the 01:37:36.500 --> 01:37:39.680 data plumbers abstracted through the data model. 01:37:39.960 --> 01:37:42.040 And so you don't end up, you end up with something 01:37:42.540 --> 01:37:43.580 that is actually maintainable 01:37:44.540 --> 01:37:47.660 and where the AI has a far higher success rate 01:37:47.840 --> 01:37:50.220 of putting something together that actually works. 01:37:50.620 --> 01:37:53.760 Then if you have to wire it all up from a bunch of components 01:37:54.060 --> 01:37:55.880 and you end up with like 10,000 lines of code 01:37:56.300 --> 01:37:58.300 that if you go back to it a month later 01:37:58.860 --> 01:37:59.780 and try to change something, 01:38:00.280 --> 01:38:02.460 it's gonna quickly get out of hand, et cetera. 01:38:02.720 --> 01:38:04.559 So this is really, we really see this 01:38:04.580 --> 01:38:09.160 as enabling technology for creating these applications 01:38:09.640 --> 01:38:13.500 quickly and efficiently and also safely with AI. 01:38:13.880 --> 01:38:15.180 I think it's a great idea. 01:38:15.420 --> 01:38:18.920 And so many people talk about AI hallucinations 01:38:19.300 --> 01:38:21.540 and AI just creating a bunch of slop and junk. 01:38:21.620 --> 01:38:23.740 And I feel a lot of that is because it's not 01:38:24.180 --> 01:38:25.460 guided and constrained. 01:38:25.460 --> 01:38:26.620 You talked about the guardrails. 01:38:26.820 --> 01:38:28.780 And instead of just saying, well, it's like Postgres, 01:38:29.080 --> 01:38:30.000 just so make me a database. 01:38:30.260 --> 01:38:30.640 You know what I mean? 01:38:30.740 --> 01:38:32.440 And it doesn't know about all these really cool features. 01:38:32.960 --> 01:38:35.880 the skill comes on, it's like, oh, yeah, you need transcripts, you need audio. 01:38:36.380 --> 01:38:39.240 It really can get a much better outcome. 01:38:39.400 --> 01:38:45.240 Okay, so the other business thing before we jump into Startup Row is I see two things in your nav. 01:38:45.620 --> 01:38:47.580 I see open source and I see pricing. 01:38:48.040 --> 01:38:50.360 What's the business model and what's open source and what's pricing? 01:38:50.680 --> 01:38:51.000 Yeah, yeah. 01:38:51.060 --> 01:38:51.620 How does this work? 01:38:51.980 --> 01:38:56.680 So Pixel Table is available as a locally pip installable package, and that is fully open source. 01:38:56.780 --> 01:38:59.260 You can just, like I said, pip install it, run it locally. 01:38:59.860 --> 01:39:04.860 And we are working on a cloud service where- 01:39:04.940 --> 01:39:07.180 so like I said, Pixel Table is a database at its core. 01:39:07.600 --> 01:39:09.680 When you install it locally, all the tables are local. 01:39:10.740 --> 01:39:13.760 There's also a piece of serving infrastructure. 01:39:14.080 --> 01:39:16.200 So you can very quickly, from a table definition, 01:39:16.480 --> 01:39:19.540 actually come up, bring up a REST endpoint, 01:39:20.260 --> 01:39:23.440 and basically serve the logic and create basically CRUD 01:39:23.560 --> 01:39:24.800 applications with it. 01:39:25.360 --> 01:39:26.660 But that's all on your local machine. 01:39:27.160 --> 01:39:28.380 Or if you want to run it in the cloud, 01:39:28.560 --> 01:39:29.380 You have to host it yourself. 01:39:30.120 --> 01:39:42.360 And we're soon going to be able to offer a basically cloud hosted tables, basically a cloud service that allow you to do all of that in the cloud, kind of like you think of an RDS or, you know, a snowflake or something like that. 01:39:42.730 --> 01:39:45.520 But now for multimodal AI applications. 01:39:46.470 --> 01:39:47.440 So that's amazing. 01:39:47.840 --> 01:39:48.460 Yeah, exactly. 01:39:48.840 --> 01:39:51.280 I think that's a really good solid model. 01:39:51.530 --> 01:39:57.680 You know, like a lot of people use these tools, but then they're just not DevOps folks or they don't want to run servers. 01:39:58.180 --> 01:40:00.440 You know, they just like, you guys just handle that for us, right? 01:40:00.520 --> 01:40:04.540 And that's a perfect synergy without misaligned incentives and stuff. 01:40:05.240 --> 01:40:06.500 Let's talk Startup Row. 01:40:06.920 --> 01:40:08.580 Why do you guys, why do you apply to it? 01:40:08.880 --> 01:40:09.880 Why do you go to this? 01:40:10.880 --> 01:40:13.520 Yeah, obviously, you know, PyCon, a very large conference, 01:40:14.440 --> 01:40:18.600 draws a, you know, a technical and very interested audience. 01:40:19.340 --> 01:40:22.560 And so, and we're obviously part of the Python ecosystem, right? 01:40:22.610 --> 01:40:25.380 It was sort of Pixel Table is fully in Python. 01:40:26.040 --> 01:40:32.720 And so this was a great opportunity for us to basically, you know, meet a bunch of folks. 01:40:33.280 --> 01:40:39.100 And we had a very large number of interesting conversations where people basically came by and said, hey, what is Pixel Table? 01:40:39.200 --> 01:40:43.520 And then very often the response was, oh, this is really interesting. 01:40:44.060 --> 01:40:50.240 I know either I am working on something related or I know a colleague of mine who was just trying to solve a similar problem, etc. 01:40:50.480 --> 01:40:54.560 So there were a lot of really interesting conversations that started that way. 01:40:54.820 --> 01:40:55.620 Yeah, that's great. 01:40:56.020 --> 01:40:59.340 I, you always hear that people say, oh, you should talk to customers as you're getting 01:40:59.410 --> 01:41:02.740 your company up and running and people who are using your product and so on. 01:41:02.880 --> 01:41:04.360 And it's one thing to put up a website. 01:41:04.680 --> 01:41:08.020 It's another to get people engaged enough to spend some time talking with you. 01:41:08.030 --> 01:41:11.540 And especially having that first experience, you know, people who look at your site and 01:41:11.680 --> 01:41:12.620 don't make sense of it, just leave. 01:41:12.730 --> 01:41:14.340 They don't take the time to talk to you. 01:41:14.580 --> 01:41:18.860 But at the booth, people come by and they, they maybe have no idea and you can get these 01:41:19.020 --> 01:41:19.500 first impressions. 01:41:19.830 --> 01:41:23.520 And so did you learn some interesting stuff talking to people at the show? 01:41:24.140 --> 01:41:32.420 Yeah, I mean, there's actually a large number of people who are building applications like this and sort of struggling to put it together. 01:41:32.660 --> 01:41:37.100 So there's a fair amount of interest in multimodal AI. 01:41:37.540 --> 01:41:43.360 And then also, as I mentioned, everybody's using AI for coding today. 01:41:43.800 --> 01:41:49.640 So being AI-friendly, coding agent-friendly is also very important. 01:41:50.000 --> 01:41:50.780 Yeah, of course. 01:41:51.220 --> 01:41:52.560 So how'd StartupRow go? 01:41:52.900 --> 01:41:53.920 You happy you went there? 01:41:54.220 --> 01:41:59.720 it's worth your time and yeah yeah absolutely absolutely we had a good uh two days i was there 01:41:59.880 --> 01:42:05.720 for two days and so yeah we had a lot of uh good conversations and not all of them i've had the 01:42:05.880 --> 01:42:11.120 chance to follow up on yet but uh we will we will do that as well so yeah we're we're hoping we're 01:42:11.120 --> 01:42:18.540 hoping it'll um it'll it'll help our business yeah i have a bunch of partnership follow-ups and other 01:42:18.880 --> 01:42:23.920 things i want to reach out to people i met at pycon a couple weeks ago and i'm still just digging 01:42:23.940 --> 01:42:28.720 backlog of email and work that I got from being away for a week. So it takes some time. 01:42:29.010 --> 01:42:29.340 It does. 01:42:29.660 --> 01:42:36.560 It does. So tell us about the tech behind this. You mentioned Postgres already. So I guess it's 01:42:36.670 --> 01:42:41.420 built kind of on top of Postgres, but tell us about some of the tech that makes Pixel Table go. 01:42:41.980 --> 01:42:47.060 Yeah, exactly. So Postgres is a component of the underlying stack. Like I said, we use Postgres 01:42:47.130 --> 01:42:53.540 for structured storage and for transactional updates to structured storage or structured data. 01:42:54.080 --> 01:42:59.620 And then obviously, because we're dealing with a lot of media data, we also need to maintain files. 01:43:00.900 --> 01:43:09.160 There is an execution engine, which basically given a query or an update, like an insert, insert, update, delete, creates a plan, an execution plan. 01:43:09.200 --> 01:43:12.040 So it looks very much like a database system on the inside. 01:43:12.620 --> 01:43:16.440 So there is a catalog that records the metadata persistently. 01:43:16.900 --> 01:43:28.140 We also put that into Postgres and then an execution system that creates a plan for any, like I said, query update statement and then runs through the plan. 01:43:28.370 --> 01:43:33.920 And part of it is, like I said, our own asynchronous execution engine. 01:43:34.370 --> 01:43:43.860 And then we have a whole bunch of integrations with external API providers like Anthropic, OpenAI, and then also integrations with things like the pill package. 01:43:44.480 --> 01:43:57.980 So, like I said, image is a column type and you can now run all the pill image transformations on your image data simply via basically putting that into a query or a computed column. 01:43:58.180 --> 01:44:02.680 Right. So you don't have to. And again, the output then is stored back in the right place. 01:44:02.920 --> 01:44:07.840 So you don't have to think about intermediate data and where it all lands or how you find it again and so forth. 01:44:08.000 --> 01:44:14.960 So we're really tying together an execution system with the storage system and doing it all transactionally. 01:44:15.100 --> 01:44:22.000 So it's a full multi-user database and with transactional, you know, isolation, atomicity, semantics and so forth. 01:44:22.480 --> 01:44:27.000 Sounds awesome. So where's things going? How long has this been around to start, I guess? 01:44:27.400 --> 01:44:28.780 When did you make it public? 01:44:29.010 --> 01:44:31.840 We incorporated in April of 24. 01:44:32.270 --> 01:44:32.400 Yeah. 01:44:32.730 --> 01:44:34.280 So a little over two years. 01:44:34.820 --> 01:44:41.540 And we're about to launch the, like I said, service for these cloud hosted tables. 01:44:41.650 --> 01:44:43.600 So basically taking a local pixel table. 01:44:44.090 --> 01:44:48.560 And then we're also right from the very beginning, it's been very important for us to give you 01:44:48.820 --> 01:44:49.700 basically a hybrid experience. 01:44:50.150 --> 01:44:54.180 And so it was always the plan that the local SDK would be there. 01:44:54.900 --> 01:44:57.240 and there would be a cloud service to complement it. 01:44:57.580 --> 01:45:00.820 It was never the idea that there would only be a cloud service, right? 01:45:01.060 --> 01:45:04.380 So a lot of, especially data scientists like to work locally, 01:45:04.900 --> 01:45:08.600 but we also see that as an advantage now with agentic coding 01:45:09.140 --> 01:45:11.760 that agents AI likes to iterate quickly 01:45:12.300 --> 01:45:14.860 and doing that locally is quick by definition, right? 01:45:14.940 --> 01:45:16.580 You avoid server round trips and all that stuff. 01:45:16.820 --> 01:45:22.260 So we're really happy that that's how the setup works. 01:45:22.420 --> 01:45:31.300 So, but yeah, so the next step, like I said, in the next few months, we'll have the cloud service up and running. 01:45:31.500 --> 01:45:31.960 Oh, excellent. 01:45:32.250 --> 01:45:40.960 If somebody is using Postgres today and they think, oh, some of these extra features, like a document column type or image column type or whatever, sounds really great. 01:45:41.500 --> 01:45:46.840 What's the process from going from like a regular Postgres setup to yours? 01:45:47.040 --> 01:45:53.480 Like, could we use SQLAlchemy to talk to Pixel Table, or do we need a special driver to speak to it? 01:45:53.680 --> 01:45:56.180 Yeah, so Pixel Table does not speak SQL. 01:45:56.620 --> 01:46:00.700 So, and Pixel Table is also not a, it's a separate database system. 01:46:00.940 --> 01:46:06.180 Really, we use Postgres under the covers, but it's not a Postgres layer or anything like that or a plugin. 01:46:06.740 --> 01:46:09.880 So you would need to ingest your data from Postgres into Pixel Table. 01:46:09.980 --> 01:46:16.680 Right. Kind of a whole two data models in memory at once for a moment and do a migration and then 01:46:16.800 --> 01:46:20.420 then just swap or something like that. Yeah, exactly. And then, like I said, 01:46:20.560 --> 01:46:26.780 it's not a, we don't have SQL support. It's not meant for analytic applications really. And so 01:46:27.140 --> 01:46:33.060 you would be using the pixel table SDK for, you know, to both express the table structure and then 01:46:33.220 --> 01:46:39.020 also create your, like I said, there is a very simple way of creating services given table 01:46:39.040 --> 01:46:44.560 definitions. Yeah, excellent. Okay, so let's close out this segment by having you just speak to people 01:46:44.590 --> 01:46:49.400 who this sounds interesting to. What should they do to get started? Yeah, I mean, my advice is 01:46:51.040 --> 01:46:56.700 look in the docs, go to the GitHub page, look through the readme, have install it. We have a, 01:46:56.980 --> 01:47:03.220 you know, a quick start guide, and then there's a whole bunch of tutorials. And one of them is 01:47:03.240 --> 01:47:09.020 probably going to reflect something that you're interested in, such as, let's say, creating an 01:47:09.580 --> 01:47:14.680 index of audio transcripts of your videos, as an example, right? That's something you can actually 01:47:14.920 --> 01:47:20.760 do in probably half an hour. And then play around with it. Okay, excellent. Marcel, thank you for 01:47:20.790 --> 01:47:25.000 being on the show. And Pixel Table looks like a really cool product. Congratulations. Michael, 01:47:25.320 --> 01:47:30.620 thank you for having me again. You bet. Bye-bye. This has been another episode of Talk Python to 01:47:30.620 --> 01:47:31.720 Thank you to our sponsors. 01:47:31.930 --> 01:47:33.220 Be sure to check out what they're offering. 01:47:33.410 --> 01:47:34.800 It really helps support the show. 01:47:35.360 --> 01:47:38.280 What if your AI agents worked like FastAPI microservices, 01:47:38.940 --> 01:47:41.220 typed, autonomous, and discovering each other at runtime? 01:47:41.920 --> 01:47:43.580 That's the world Agent Field is building. 01:47:44.200 --> 01:47:47.100 Join them at talkpython.fm/agentfield. 01:47:47.560 --> 01:47:49.440 If you or your team needs to learn Python, 01:47:49.650 --> 01:47:53.060 we have over 270 hours of beginner and advanced courses 01:47:53.250 --> 01:47:56.800 on topics ranging from complete beginners to async code, 01:47:56.940 --> 01:47:59.720 Flask, Django, HTMX, and even LLMs. 01:47:59.920 --> 01:48:02.220 Best of all, there's no subscription in sight. 01:48:02.880 --> 01:48:04.540 Browse the catalog at talkpython.fm. 01:48:05.280 --> 01:48:08.600 And if you're not already subscribed to the show on your favorite podcast player, 01:48:09.260 --> 01:48:09.900 what are you waiting for? 01:48:10.580 --> 01:48:12.340 Just search for Python in your podcast player. 01:48:12.450 --> 01:48:13.320 We should be right at the top. 01:48:13.740 --> 01:48:16.500 If you enjoyed that geeky rap song, you can download the full track. 01:48:16.730 --> 01:48:18.620 The link is actually in your podcast blur show notes. 01:48:19.420 --> 01:48:20.780 This is your host, Michael Kennedy. 01:48:21.180 --> 01:48:22.240 Thank you so much for listening. 01:48:22.470 --> 01:48:23.240 I really appreciate it. 01:48:23.680 --> 01:48:24.380 I'll see you next time. 01:48:36.260 --> 01:48:39.060 I'm out.