Bay.Area.AI: Interview with Roie Schwaber-Cohen, Pinecone
Recording: Bay.Area.AI: Interview with Roie Schwaber-Cohen, Pinecone
hello everybody my name is Alexi kov I'm the founder and organizer of Baya AI which is the most established AI met up in Bay Area and the world and here we are with Roy who is theel Advocate from Pine con and we later have a program on rag in various kinds Vector rag graph Rag and uh I'm a Comm architect at new for J and we have an integration coming up with pine con and uh the there is a blog post on both NE J and pine con so we're very happy to have Roy and we're just going to talk about Dev things yeah yeah yeah sounds great so tell us a little bit you know what you do at Pine Cone how do you do the devel and where have you been before like how you came to this point in your career yeah for sure I'll start with the the second part um I've been an engineer for almost 20 years um mostly worked in startups um and started at Pine con about two years ago after doing my first um devant at a company called the serto out of Seattle for that worked at a company called Mana that dealt with um a lot of AI before the chat GPT craze started um so at at um pine cone I would say that like the job is kind of like broken into two two main parts one is kind of ideating and figuring out interesting use cases for our customers and kind of like showcasing the capabilities of the product right like and exploring different modalities um that it could be used in um and because um Pine con as a as a database a vector database um it's like a very fundamental piece of a lot of different AI applications and so we have like the leeway to kind of decide where we want to go how we want to attack um which vertical um and you know for as a Dev like it's a lot of fun um to to have that like you know clear canvas to kind of experiment on interesting I like that like you know you mentioned that you were doing AI before eii become fashionable right like we started at arop 10 years ago with NLP and then data mining if people remember that machine learning and now it's all AI right so I wonder like you've seen AI evolve into this do you see any patterns like the first wave of deep learning like because I think a lot of folks just came into this and the thing is all new like do you see any he's repeating yourself any kind of fun parallels um yeah I mean I would say that like you know there's always like the hype cycle and release of a lot of different technologies that that we've seen out there um and the question is like always like what's the staying power so similar to mobile and social that kind of rose and kind of became a thing right and you don't talk about them anymore even though they're still powering so much of our Lives um I think AI is going to become that you know so out out of the hype cycle perhaps but still very like Central in our lives in ways that are a little bit more subtle and in that sense like I think that what happens with a lot of Technologies um in this space is that again they stop becoming like this one shiny thing and instead people start understanding how to compose them with other existing Technologies and capabilities with AI I think it's going to be the same so instead of you know just thinking that the entire word is llms and only llms are going to save us I think that with Rag and other patterns are sort of developing we're so slowly beginning to understand that like there's a lot of value still to be had from systems that we would call traditional or less modern quote unquote um and that's that's exciting to see that's awesome yeah I agree with you it's like you know there is a whole stack right and there is like emerging Stacks I wonder if you can talk about like Pine was the thing a leader in R right like you guys were early out of the gate uh so how do you kind of how did you see it from with inside the evolution like what what are the parts of uh AI stack you see the most like how do you see this evolving like with pine cone in the middle of it yeah I think that like the fundamental understanding on on in Pine Cone and spe specifically our founder Ido Liberty Liberty um is that you know know the the volumes of unstructured data are going to rise and that the best way to to leverage that data would be in the way of embeddings right in the way of some some Mech when we say embeddings we say some mechanism that can make sense of this unstructured data in some programmatic way MH um and I think that in that sense like it's a long-term BET right so like there's all of these different modalities that we're seeing now right like rag is one way to leverage that capability of handling unstructured data by using embeddings yep um you know doing image search and doing all sorts of other kind of um operations on structured data on unstructured data can prove to be um very lucrative as well in the future right so I think we're keeping our options open but we're also trying to leverage on opportunities that are kind of immediate where we can help our customers and our users by making their lives easier um for example with with rag right like which is which is still a very big portion of um of our of our activity um we just released um assistant and that is just a very simple way of getting getting a rag stack kind of started um without having to know the intricacies of you know how to do chunking and how to do embedding properly and how to like go about all the evaluation steps on your own you get that as as as a sort of like an API um and I think that is as a direction right like we have this like very wide scope of things that you could do with the infrastructure but also very specific type of implementations that we want to kind of Usher and help people build that's awesome I I love it right so like assistant will help people to take their documents put them in Rag and do it properly and then like they're like go they own business they in business that's great so so you're at the and like you go to deel like you talk to developers can you talk a little bit like what do you see because it's super hard like you know I think the difference from previous waves of AI like you think de learn was was was hard like look at this stuff like the the the change is much much faster there's much more new stuff right like how do you cope with this yourself and like how do you see developers coping like what questions do they ask the most like how do we go about basically teaching the community the best practices yeah it's an excellent question so um I think that like you're right like I think that the people are overwhelmed by like just the sheer number of possibilities in front of them mhm um you know there's I I would classify like you know the world into two very general groups like people who know what they want to do and people who don't know what they want to do so and I think that there's like a a very big group of people who see a lot of very cool tools and are like I want to use these tools but they don't know what they want to do with these tools these people have like a very different problem from the second group um which we kind of try to empower right so like the first group right the our job as devell is to basically show The Art of the possible right so like basically show them what are the things what are the types of problems that you can go about solving with this thing and then once you've found how your problem maps to this problem you're more than likely to follow the same path right so these are the materials we'll pull out for them for people who know exactly what it is that they want to do they just want to get to the Brass tax of how to stand these things up right like the problem there becomes more of knowing all of the different details and fact factors that that that that influence the way you construct your stack as a whole um that gets like the combinatorics of that kind of explode now when you have like all of these new things and systems that you have to think about that you have you need to think about before M um so part of it is kind of like just general engineering um premises that are for example right like when you build the rag P pipeline MH don't just build the rag Pipeline and say things will be good you need to have an evaluation right like Paradigm tool mechanism there in place to actually know right that you're doing a good job we not doing good job so like even going about that whole pipeline is not necessarily true for a lot of people so that's the other aspect right like it's trying to give people the tools and the and the and the flows right and examples that kind of tell them how to compose all the different tools that exist in this ecosystem because unfortunately for us right like we we cannot as a database right operate in like a vacuum like we're reliant on the entire like one right like and then delegate right exactly that's great and so uh the last question for the ra and then I just have like a fun question uh is we started the new series called D not a right and so that's also URL of course and this is basically how to keep it real doing the RA in the times and so you've been doing it for a long time I'm coming back to theel you know after like interesting things so uh what's your advice to devels operating in this super hard world how do you keep it real how do you make it interesting what should you do to really like make developers happy yeah I mean for me at least the The Joy has always been at helping people understand what problems they're trying to solve specifically to sort of try and untie them from the ways that they think about solving a problem and you know that becoming kind of like the the the thing that cor Corners you into a very particular solution or even worse right when you start with a solution and like then you limit the scope of your problem to begin with right so I think like that like having like the knowledge right um of the tools that you're that you're uh promoting right is super important because then you can enable people to think about problems in a certain way right and I think that people that work with you know graph databases have a very particular way of thinking about the world I think that now with my experience using vectors and graphs and other things like type systems Etc have a way to think about the world um uh and I think that as long as I can you know help people I understand like what their goals are very specifically right then it allows me to you know come up with potential you know recommendations and Technical Solutions and actually be effective so I think that that works for me I don't know if it works for everybody I love it and I think it's like the it's it's you know it's the like Eternal truth know your tools learn your options right like it's it's not magic learn how this stuff works you know programming language know you know it well know EV Val right like learn the components and then you can put them together y love it and the last question like tell us something fun about yourself like what do you you know what's a fun fact about you what do you like to do for fun yeah um the fun facts I'm I've been playing flute for about 25 years years or so oh wow I play other instruments is it classical music uh I don't play classical music anymore I used to play classical music I used to be in a choir when it was like a really fantastic really young kid um but yeah no I'm still I I still love playing music I play piano bass stuff like that um but yeah mostly with myself fantastic this is great like you know I think a lot of folks in AI you know originally started with artists philosophers musicians like it's really great to have all of this coming together well thank you Ro so much we're looking forward to your talk all right thank you so much all right