Transcript: Ankush Gola, LangChain — Interview with Alexy

Hello everybody, I'm Alex Krarov from Lakes Hall here on location with Ankush who is the CTO and co-owner of Lakes Hall in their new office. Have they been here 3 weeks? >> Yeah, we've been here for 3 weeks. Yeah, great to have you Alexi. >> So you're hosting the Datafusion with with Hadera. Tell us a little bit like what's your use of Datafusion? How come you guys picked this? What's like your due to Rust? And all these good things. >> Yeah, so we use Datafusion to build our data layer for agent observability. It's called Smithy B. It powers our agent observability product Linksmith. We decided to use Datafusion because it just came with a lot of capabilities out of the box. We didn't have to reinvent like query planning or optimizer rules or anything like that. It gave us the flexibility to make our own execution plans and optimizer rules and our own dialect for querying as well. So yeah, it was a great balance of like batteries included and extensibility. And yeah, it just had a great community behind it. So super excited to support that today with our meet up. >> And I just talked with you a little bit before. So again, I'm learning about 3 DB and you think I'm an expert. So you mentioned it's basically a novel kind of database. It's supporting your own product. It's meant to speed things up and basically make agents more efficient. >> Yeah, agent observability more efficient. Yeah, so basically the problem that Smithy B solves is like agent observability is a pretty challenging data infrastructure problem. The payloads associated with agent traces are large. The types of queries that you have to perform on them are you know pretty bespoke I would say as well, pretty unique. You have to be good at like random access queries, like time range scans. You have to be good at analytics. And so we had to build something that had the right balance of performance, like flexibility, portability for our specific use case. >> Yeah, and so how do you go like do you program with Rust or do you use Datafusion out of the box like >> Yeah, we are we are we are we are fully like Smithy B is fully built in Rust. Yeah, that's how we're able to leverage data fusion. >> yourself Rust developers or do you just go like tell your developers like just like go >> I I no, absolutely not. [laughter] No, you cannot write code in a database. Uh Um no, yeah. So, I'm I'm pretty hands-on. I've done a lot of the SmithDB development. We have an awesome team here um of uh nine people and you know, we're all like super hands-on Rust developers and you know, trying to make this thing better and better. >> So, this is exciting because you know, I started the community in July at AWS uh Builder Loft and we really hope you guys will come and speak and my promise was that AI is now basically Python on top and Rust uh beneath. It's like an iceberg. The tip of it is in Python and infra is in Rust. Do you think this is the future? This is how we're converging? >> Yeah, I mean I I I think like for certain systems it definitely makes a lot of sense to develop in Rust. Um in some systems it makes less sense to develop in Rust. Uh but yeah, I mean I think uh we actually started built you know, LinkSmith was uh there's a lot of Python in LinkSmith. We started moving some stuff to Go. We moved the entire data layer to Rust uh with SmithDB um and so yeah, I think it's a common pattern that a lot of people take. >> So, super excited. Thanks for hosting. We're looking forward to talking to you. >> Appreciate it.