Prashanth Rao, LanceDB, on Reliable AI — Interview with Alexy

Yeah. Hi, my name is Prashant. I work at Yeah. Hi, my name is Prashant. I work at Lance DB as an AI engineer. So, I've Lance DB as an AI engineer. So, I've been working in AI for a few years now. been working in AI for a few years now. And for me, the biggest highlight of And for me, the biggest highlight of what I've been doing is uh seeing the what I've been doing is uh seeing the work that I do uh mean something to work that I do uh mean something to people when I release something in open people when I release something in open source. And the beauty of the current AI source. And the beauty of the current AI ecosystem is that so many amazing tools ecosystem is that so many amazing tools and frameworks are open source. So the and frameworks are open source. So the sharing of ideas and the stuff that I've sharing of ideas and the stuff that I've done and feedback from the community is done and feedback from the community is what I take away from the past few what I take away from the past few years. So AI by nature is very years. So AI by nature is very non-deterministic. So I think the real non-deterministic. So I think the real challenge in today's AI ecosystem is challenge in today's AI ecosystem is building systems in production that uh building systems in production that uh produce results relatively uh produce results relatively uh consistently and reliably and graceful consistently and reliably and graceful handling of failures and I think a lot handling of failures and I think a lot of engineering work is needed in any AI of engineering work is needed in any AI system. So reliability means a system. So reliability means a combination of all these factors so that combination of all these factors so that systems run smoothly in production. So I systems run smoothly in production. So I think there's a lot of need for good think there's a lot of need for good tooling in a lot of the layers that are tooling in a lot of the layers that are peripheral to the language models. Um peripheral to the language models. Um the the AI models basically produce the the AI models basically produce outputs but the scaffolding around that outputs but the scaffolding around that including things like observability including things like observability guard rails. Uh there's a lot of tooling guard rails. Uh there's a lot of tooling around that that I think is is maturing around that that I think is is maturing as we speak and I think the ecosystem is as we speak and I think the ecosystem is actually really evolving very rapidly. actually really evolving very rapidly. So I think the community is also So I think the community is also learning these concepts as they build learning these concepts as they build along these lines. So I think the AI along these lines. So I think the AI stack is going to rely on many open stack is going to rely on many open formats. I think format specifications formats. I think format specifications are a really big deal. Uh data formats, are a really big deal. Uh data formats, uh table formats, I think a lot of other uh table formats, I think a lot of other uh infrastructure uh at the low level of uh infrastructure uh at the low level of the stack and a lot of it is related to the stack and a lot of it is related to scalability because AI is generating scalability because AI is generating data at a breakneck pace. So I think data at a breakneck pace. So I think there's going to be a lot of innovation there's going to be a lot of innovation um on the storage side and scaling to um on the storage side and scaling to bigger and bigger data sets.