Transcript
Mike Matchett: Hi Mike with Small World Big Data. I'm here today talking about, I don't know, several of my most favorite things! We're going to have AI in the conversation. We're going to be talking about the software development process. We're going to be talking about coding and getting answers. We're talking about data science and how to get the most value out of your data and bringing that all together. We've got Zerve here today, so just hang up. Hey, welcome to our show. How are you doing today? Phily Hayes, CEO: Doing really well. Excited to talk. Mike. Mike Matchett: Uh, just just briefly tell me a little bit about how you got involved in bringing AI into data science yourself. Like, what was what sort of your background or interest here? Phily Hayes, CEO: Yeah. So my background was really a co-founder, um, that I started with, uh, as I said, one of the smartest humans I've ever had the chance of encountering. Um, he was working post his PhD in the data science space and was really struggling to get those projects over the line. And I thought, well, personally, if he couldn't do it, no one can. So I thought it was a mission worth joining, and that's what got us started in the space. Mike working on it. Mike Matchett: All right. And was this something that you started working on before everyone was getting really excited about AI, or was it the AI opportunity that you said, hey, I can put these blocks together and make something here? Phily Hayes, CEO: No, we were working on this before. We had done hundreds and hundreds of hours of interviews starting from kind of mid-May 2021. Uh, working with what we refer to as code first data teams and code first data users trying to understand what were their biggest pains and problems in getting data science and AI projects as they existed then into production. And largely a lot of those problems, they may have changed and shifted, but a lot of those problems that came through into the kind of Lem world that we exist in today. So it positioned us really well to tackle the set of problems that we are now. Mike Matchett: Right. And just for, you know, those of you watching, this is a little bit of a meta conversation, too, because we're going to be talking about building data science code to create possibly machine learning type activities and services that would then also be used by the AI. While we're using AI to do it. So I hope we can keep up, but we'll do our best here in the next few minutes to make this clear. So let's start a little bit. This is, you know, you're really tackling these data science complexities and challenges that people were having just briefly, you know, what were you seeing as sort of the main issues that people in the data science world were having is just complexity? Is it repetition? Is it the constant pace of change, the growing size of data? What, what what sort of landed for you? Phily Hayes, CEO: Yeah. So it's slightly evolved. So I think the core thing that stayed the same as we've came through in the journey was getting things into production and making an impact. These, these teams and users and people weren't making the impacts we felt they could have been. And so that was that was core to what we wanted to solve in the LLM world. The yeah, the rate of change is so unbelievably quick. There's new frameworks, new models. You need things to be unbelievably open and flexible. And that was core to us building an operating system that allowed them to do what they do best. Work through code like be an expert in their field, but have abstracted away some of the software development components. Have abstracted away some of the cloud computing resources, as well as abstracted away how to connect to either open or closed source models in order to make impactful data and AI projects. Mike Matchett: I know this is tough to do without actually visually walking through the product a little bit, but just just at a at a high level. What does it look like? Sort of before and after? If someone's doing a data science project and or an initiative or they've given a task and what do they go through before they start using Zerve, and what does it look like afterwards when they're using Zerve? Phily Hayes, CEO: Yeah. So I guess before there's a whole host of ways you can start this, there's been the kind of low code, no code movement. There's people that are working through code and in notebooks and vs code. And then there's lots of large, I guess, enterprise players, enterprise AI players that you may be working through. And each of those has their own set of complexities and issues. Some of them still boil down to the same things. It's very difficult to make an impactful projects and get that the whole way through into production. So really at its core. Um, you know, there's differences, but a lot of that still stays the same in terms of the core problem. From a perspective, what we've tried to do is create, you know, um, essentially as a starting point, your playground to start working with data, to start working with your large language models, all in an enterprise like okay zone. So a place that has gotten through your enterprise, it great solutions. So that allows you to start working through a directed acyclic graph. So we build a graph on our, uh, on your canvas. And that is a coding environment where you can code in each one of the blocks and the blocks as they go from left to right, depict your both your data flow as well as the orchestration of how that will run. So you're really kind of visualizing something you're working in, something that I think a lot of people are used to, especially in the data engineering space, building out a pipeline, and you're doing that through code snippets. What's happened as of late is utilizing agents to start building that as well. So you've got an agent on your team now that, of course, has access to the code and is able to write code, which has now became table stakes, because ours is built specifically for working with data and building AI. Ours also has access to that data, which is amazing context in order to build these sorts of projects. And because it's able to create those blog, those blocks, rather it's creating infrastructure, it's running that infrastructure. If it encounters an error, it's going to try and self-heal just like a human would. So it's really multiplying the level of output that these teams have had. So it's a, you know, a huge contrast in terms of the before and after. And I'm like. Mike Matchett: Yeah, I've seen a little bit of it. And what's interesting to me is, as you're saying, and I think it's worth repeating for someone watching this, is that it's not simply helping you write code or giving you code suggestions. It's it's encapsulating the operation, like the DevOps essentially, of that code block as well. So you've given it some parameters or said, here's my environment and it's it's saying, okay, here's what, here's the code that needs to run and here's how we're going to run it, and when we're going to run it, how we run it. And then here's here's that graph of how it connects to the next part of the pipeline. So it really takes on a number of jobs. This agent really for the folks that are doing this, what kind of what kind of speedups are you seeing in terms of productivity when people do this. Phily Hayes, CEO: Yeah. So so huge speedups like we've seen people go from prototype to production nine times as fast. We're seeing teams take on 8 to 10 times as many projects as they were in terms of capacity. So just huge amount of speed ups across the board. The other part that is causing is having a massive impact is the abstracting away of software development skill sets. So if you've got a team that was really well able to handle model build algorithms for internal applications, but they weren't necessarily building tools that needed to have four nines or five nines of uptime. That's a lot of what we've tried to abstract away. So we've already tied in your CI, CD and software development lifecycle. We've tied in all the Docker containerization, logging, monitoring so that those skill sets aren't things you needed to add so that your data, AI team and builders are getting things actually into production from that perspective as well. Mike Matchett: All right. This sounds a little bit like it's a big deal to deploy or use or get up to speed with, but I understand it's not that hard to tell us what this looks like when I go to deploy this and bring this into an organization. Phily Hayes, CEO: Yeah, super straightforward as well. So hard for us to build, but thankfully easy for our users to use then. So for example, we're across all the clouds on prem. But if for example, if you're on AWS, it's one click and within six minutes we're sitting inside the VPC of the of the end user. So it's sitting inside their cloud environment. So none of the data is leaving their their infrastructure, which was really important in terms of starting from that perspective and then starting from a user's perspective. There's numerous ways you can drag and drop a Jupyter notebook onto the Dag. It will automatically build that Dag for you, which is a super cool feature, but also the agent just changes how you start. So the so from a starting point, it's it's so straightforward to say, okay, I'm thinking about doing this. Let's start that process. And the agent is going to in front of your eyes, start building it. And what's really interesting about that is depending on the prompt, more than one agent may be used to to split up that tasks. So just like you might be creating a parallelized computing workflow, you might actually have parallel agents working on that as well. And you're working in-line and watching it. So we fundamentally believe it's the perfect interface for human agent collaboration. And it's so super straightforward to get up and running in your infrastructure, but also start the project because of being able to interact with the agent. Mike Matchett: So if I've caught that, I can take a Jupyter project that I've actually demonstrated to somebody, drag it in there and get going right from the start, and it's going to implement that in a production oriented way for me and take care of all that stuff. I mean, that's pretty cool for some people already without talking about like how it helps you make up stuff and make APIs and, and because you can build all sorts of different kinds of things just real quickly. Right? Phily Hayes, CEO: Yeah, absolutely. So we have people building applications. We have people building APIs scheduled jobs. And as you alluded to earlier, agents, the agent is now helping build agents. So it's very meta from that perspective. Indeed. Yeah. Mike Matchett: All right. So if someone I know we don't have a whole lot of time here. There's a lot to cover. I hope you guys come back. But if someone wants to kick the tires. Here, take a look. Dig a little bit more. And I think anybody with a data science team should take a look, because this is really looks like a huge accelerator that. Phily Hayes, CEO: We'd love to have them. So there's an absolutely free community tier where they can go in, kick the tires, try the agent, try the Jupyter notebook, uh, import, try building, uh, building agents, apps, APIs, all of those things. All on the community tier. Um, yeah. Any feedback? Join the slack community. Um, they can ping me on my calendar for a zoom coffee. Happy to take the feedback in person. So, yeah, would love to hear what they do and see what they build. Mike Matchett: All right. Thanks for being here today. Thanks for explaining. Phily Hayes, CEO: Thank you for your time. Mike Matchett: And come back. Uh. That Zerve. Check it out, folks. If you've got data science projects, initiatives, trying to do something, it's well worth your investment to take a look. Take care.
