Sri Desikan, Elastic & Vrashank Jain, Dell | Cube Conversation

Hi, I'm John F with the cube here in our PaloAlto studios for cube conversation with Dell AI data platform with elastic. We got two expert Shagir and Shri with Dell and Elastic. Two experts in the data platform. They've seen the movie. They've seen multiple ways of innovation. Guys, we are in a perfect storm for innovation when it comes to data. If you're in the data business, it's almost like, man, I wish they had this 10 years ago. >> Um, all the mechanisms to manage large systems are getting having intelligence injected into them. They're being figured out and abstracted away from users so they can free up for more time. More data sets are coming in, >> more data types. Now you have AI models, frontier and specialized models. So you have intelligence coming into the data business through these models and high performance computing, AI factories, AI at the edge. We're going to have a lot of horsepower. >> Yeah. >> What does that do to the data industry? Because you have data scientists, you got data engineers, you got analytics folks that have been building dashboards. They're used to certain things. Now you got the DevOps side. They love love setting up these large platforms. Shift left for developers. Yeah. >> All this comes together. What's the biggest change when you inject intelligence in? >> Oh yeah, there's um I think it's transforming how people work. You know, there's a lot of talk about whether this is going to take off take away jobs and so on. What what I'm seeing personally is um there's a there's a lot of um you know enhancement that these tools give to existing uh knowledge workers. So let's take data science as an example, right? What is the role of a data scientist? Now if you really look at different models and the data evals or evaluations are the most important thing for you to know whether an agent is going to be accurate or not and that eval is performed by data scientists. So the role has become different right so it's not eliminated it's different the same thing with um observability we have sur agents but then you have human in the loop and the human in the loop has to be part of the conversation with the agents so it's like the s sur having 10 more people or interns being able to do the job 10x better right so I feel it's more of a transformation and not elimination right and then and I I think that's going to get even better with other knowledge work as well I I actually think like we haven't even seen the best of deal platforms yet. Uh I mean they they were all the rage in like 2020 or something. Uh and then you know chatb came along and suddenly it was generative AI for a few years. Now it's a Gent. >> I actually think the the best of data platform is yet to come and I think it's because of the fact that it's finally getting its hands on what is arguably the largest data set that has always existed which is unstructured. >> We've said this before, right? 90% of data is unstructured. It's dark. yada yada yada. We finally have a killer use case, >> we finally have the path to tapping into it >> and I think the more people lean into that side of the house, which is let's start solving the hardware problems now with with data platform. I think that's where that's where we're headed, >> you know, and I've been a um student of enterprise search going back to the 90s when when I was working on my search products back in the day. It's always hard in the enterprise. You have all this legacy systems permission >> that are actually outside the scope of the schemas of the software. So you have it was always hard to crack the code on enterprise search. Now it's you guys have done that with elastic. So you kind of hit that. Now you got the the retriever side of gen one and chat bots check the box on that. You get vectors. Now you got reasoning. >> So I think the the the conversation that's been happening on the cube is if a company could have their own brain. >> Yeah. >> That's the company's knowledge. Yeah, >> it's like going to school. You graduate and you apply more. You relearn and people who are curious, who are always learning, that seems to be the memes and best practices on Instagram and Tik Tok. This is like how we're seeing the enterprises start to think about their business. >> Okay. So, if that's like a preferred future, where how do we get from here to there? Graph databases have a good fit because neural pathways are kind of graphs. Yep. >> You know, so >> as the data will form, I think you're right. I think the best is yet to come. Yeah. >> And no matter what you throw at it, >> the brain will figure it out. The arms and legs, the nerve centers. So, you got, you know, there's a lot of like biology examples being used in AI. >> Um, systolic arrays, which I learned over the weekend, is like a bloodstream is the data. The brain is the CPU, a GPU, and the processing units are the cells. Yeah. Right. That was what they were using at Stanford this weekend, which I like because it's like, okay, but how do we get the database if the company brain >> Yeah. >> is the moat. >> Yeah. >> You just keep it fed, you keep it knowledgeable. AI >> keep it fresh. Yeah. I think that's >> the AI helps that doesn't really there's no competition to that. AI helps the brain better. So, how does a company set up their brain? If you had to kind of do it all over, what how do we get from there to there? >> This is what we've been talking about as context engineering and context layer, right? But if you really think of the paradigm for databases over the last 20 30 years relational columnar and you have graph and the question is you know data is in different shapes and sizes. So you sort of need a context database so to speak and and that database is an amalgam of so many different things time series data column data documents >> and uh that's where elastic is going in terms of capturing all of this context >> and allowing the agents to use that context at the right time and pass it to the models right so so a database sort of becomes no longer just a relational or a column database it's all of the above >> you can have everything it's like when I used to tell my teenager frontal loes not developed. You know, like you can have a brain that has different databases contributing to the overall, you know, where you are in the journey. Time series here, relational here, tons unstructured, and that's essentially this the tissue. I think you're right about that. So, I think you're going to it's not about the database, it's about how do you connect them all? >> Yeah, I I think what I think the there's going to be a few words that are going to get thrown around a lot in the next year or so, and it's going to happen. context layer, semantic layers, uh knowledge graphs, and ontologies and all these things. If you cut the noise out for a minute, basically what we're what we're trying to tell people is don't just stop at having a really good index. Don't stop at having a really good database or a view or whatever it may be. Go maybe three or four steps forward. The first step, I think, is to your point, we need a common set of definitions so that agents actually know what the heck they're talking about. when somebody says I want revenue from the past quarter I should know what quarter stands for and every company you know has a little bit different the second thing is I need to be able to say if I ask for a quarter revenue I should also be looking at margin and that lives in as a different database that should be connected somehow and therefore there's a graph which is effectively our brain but then I think we take it one step further which is and it's really imitating the way our organizations work which is when you wanted to ask a question >> and you didn't know the answer what would you You would call a guy who knew a guy >> who knew a guy. >> Those people >> phone a friend and say, "Help me figure this out." >> Exactly. That's what agents are doing now. They're phoning a friend. And so I think the next step in the data platform is now exposing data platform as a collection of really highly curated agents that are really good at a particular part of the graph. They just know the financial part of the graph really, really well. They have access to the financial contract documents. They have access to financial databases. An agent should come in and say depending on the question am I supposed to ask this guy or that guy which basically says which agent am I supposed to call on >> and that agent should also be able to call another agent. So it's >> replicating the org in a way >> in an agentic fashion. >> Shri it's almost as if you know I know you guys are in the search business but you mentioned graph multiple time but still a search paradigm under the covers. Yeah, >> the search engineers and the search practitioners, this is the time that they're the ones who have the superpower because >> all the value coming out of this connective tissue where we go to the graph, brain, database management, data management, >> that's driving revenue because you get faster thinking. >> Yeah. >> Faster thinking, faster reasoning, faster knowledge and retrieval. And I think I think part of the thing that gets lost in you know in what's important for the AI world is the there's there's the there's the human world of curating data and you know storing it as a graph and a a table and so on right but we always forget that in the non-human world in the agentic world retrieval is so important because you need to retrieve it >> and retrieve it fast because an agent is probably spawning you know five sub aents or 50 sub aents and as Rashank mentioned each one is having a particular task and you need to all of this to come back in a reasonable time otherwise no one's going to sit so um I think retrieval is as important as just >> you got to get it out of the database and into the into the mathematics of the >> and look at different data shapes and still retrieve it fast right so that's an a very important part of uh the the equation here uh so so when we talk about context it's not just about storing the context as a graph and ontologies and so on it's also being able to retrieve it really really fast at the speed at which you know agents need to be operating and and agents are part of a larger workflow typically right we we don't we haven't talked about orchestration and workflow >> so a workflow could have multiple steps and and each step could be a call to an LLM so this entire mechanism needs to work really fast with the right recall and precision and speed so >> yeah if you want to tap all your brain and be that movie like Limitless we all seen some the movie on that >> you want to tap into all aspects of the brain that's latency that's orchestration >> you don't want Wait for the brain to warm up retrieval. >> Yeah. >> Yeah. Go. I would say my left brain, right side of the brain. Go be more creative. >> This becomes interesting. You mentioned ontologies and you guys were talking earlier. We did a segment for your October 6th event which is going to have all the the right people in the room to talk about this direction. >> Yes. >> Is when I hear Palunteer talk about ontologies and everyone on Wall Street's like, "Oh yeah, it's ontologies." It's like the hottest word that's been around since 1985. Yes. >> Or actually 70s, I think, was the original. But >> it's not a new concept. >> No. But it does highlight what we were just talking about on your preview video of your event is that it's integrated. It's it just shows that you can't actually run an antology and a knowledge layer or graph or anything without being in some sort of status system. Yeah. >> Or like explain this phenomenon. It's not an operating system per se, but it's kind of like a data operating system. They have to if we're going to have a brain better know what's going on. Has orchestration kind of has scheduling if you want to think of it that way. Yeah. >> So these are computer science topics now applied to the data world. >> Yeah. >> Yeah. You mentioned SR, the whole SR fleet and of agents in our heads. >> Yeah. >> Yeah. I mean, onto like to your point, ontology isn't new. It's it's it's like a it's a combination of a standard operating procedure and a constitution all built into one, right? And we've been trying to build ontologies for decades at some point. And a lot of companies have tried to do it manually. >> Now you got intelligence and ages and coding on the fly. >> Yeah. The demand for ontology Exactly. The demand for ontologies is is only getting bigger and bigger and bigger. But uh I think what Palanteer did really well in the in the last six years or so is they really cracked the nut on how do you build robust ontologies and I think what they realized >> is that you're going to have to do it custom >> which is why a lot of palenteer engagements tend to be longunning. They tend to be deep uh you know obvious solution >> right and it's they don't promise results overnight. They say, "Look, we're going to have to sit with you for about 12 months or so, 9 months or so, and we're going to get you over the line here." But Palant >> and the benefits are pretty quantifiable. >> Exactly. Right. The benefits are obviously from a national defense perspective, you get tremendous amount of benefits. Uh but even for regular enterprises, we're seeing that. I think what's Palanteer has proven is that you needed you always needed ontologies. You could really do them well now with data. And I think finally, if we're able to crack them out on on using generative models to build it and maintain it. Yeah, >> we're able to I think spread this benefit of ontology to everyone without having to sign up for 12. >> Well, the graph database is a nice tell sign for the direction because what now said okay I can recurse through things. Another computer science word that you hear now in the data world a lot. AI models speak recursive too and you hear that a lot >> is that you can actually get more faster taxonomy builds. So you actually use the intelligence but you have the frontier models are going for the AGI that's general intelligence they they crawl the internet right >> they're not crawling the enterprise >> that's not in the general intelligence yet but now that the specialty of the enterprise >> so specialized intelligence that's why we see companies like fireworks AI doing extremely well because they're asset light y >> they're not really spending any capex they're just a layer between a specialized intelligence the open weight models Yeah, this points to the future because I can run a really great model for a thing that I know I want to run it on and I don't have to bring everything else in. >> Yeah. So you you mentioned Fireworks. So one of the things that I think is new in the enterprise in the last few months is so people are realizing that they can fine-tune post train models right and fireworks does that. Um so where does that where does data fit into it? In order to post train models you need data. >> It fits nicely into it actually. >> It fits nicely. It does. So uh the real real world data as well as being able to generate synthetic data and use that for post training is becoming people want to do that because >> so this is where I think the brain comes in enterprises actually probably have their own brain now that would be a good board pitch because hey your mode is your data >> so you want to have an autonomous organization which could be a kind of a visionary statement. Yeah. >> How do you do that? Well, it's going to come from your data >> the data world and we're seeing the new implements of some of the graph stuff with with good search um platforms >> just turn into instant value. So all the interviews I've done in the past year hits this hard. Yeah. >> People just get and they have to explain to look it's really good. You know I have the answer. >> Yeah. You know >> it's just been hard to I think it's been traditionally hard to convince people that you could actually build it and maintain it. Like the idea isn't novel. It's just that we knew just how hard it was to build an enterprise brain and actually keep it >> high quality and and worth investing in beyond the first six months. >> All right. My my final question on this little little riff here with you guys on this kind of podcast format is there's a lot of young people and systems people that are, you know, over the age of 40 that have been doing database work, ontologies. They're now back in the game. You got this fusion between young and old data pros. Yeah. >> But the young guns don't know their data pros yet. they're just AI native. So you have this kind of confluence for folks that are getting into the game on AI native. What do you think they should learn knowing all the legacy knowledge we have in the history of it knowing kind of what's been and where it goes. What advice would you give to these new young engineers that don't even know the semantics of SQL? I already kne agent I don't need to learn SQL. So they probably don't even do SQL. >> So they're they're already past the point of us saying that. So what do they need to learn to get to that that to be a contributor in the community? What would that you give them advice would you give them? >> You want to go first? >> Yeah question. >> Uh it's it's a really good question, right? Um we grapple with it with our own engineering teams like as they evolve and what are the key skills that are needed, right? So first an understanding of how models work, right? You know in terms of you know what what causes some a model to behave a certain way. So that's always good. And you mentioned young people who are not exposed to data. They start at the model and then work their way down. And if you're a person who already knows data, then I think we talked a lot about it today. What are the types of data, the data shapes that are needed in order to feed the right context at the right time to the model, right? So you sort of approach this from both ends and and there's probably a sort of a nirvana in the middle, right? the old >> it's like we use the biology example the highest paid physicians are brain surgeons heart surgeons and neurology I mean your nervous system your brain and heart I mean that's the number one thing so the brain is the models right >> your brain surgeons >> right and then and then the hands are the tools so what what what are the right set of >> robotics I just saw some great demos from unitry this weekend um yeah I mean but this kind of this is a a rhetorical question but this is what's on the mind of people how do I tune my intellect and energy to maximize. >> Yeah. I I I think my my answer would be like less technical and much more sort of I don't know philosophical but >> just continue to build your problem solving skills and just don't give up because we're seeing just I it happens to me too like I think my brain is getting lazier just because I'm expecting a model to help me out whenever I reach a certain issue and we shouldn't give that up. We shouldn't give our give up our own agency to to think creatively, thinking out of the box and only relying on a on a on a clawed model to actually tell me what else I could be doing. >> Yeah. Yeah. >> Um so just focus on >> it's like going to the gym. You want to keep that mind going, be curious, >> work out, keep that >> and there's plenty of opportunity to drive intelligence. >> Yeah. Exactly. >> The ingenuity and engineering mindsets could >> you don't actually have to know the answer. You just >> know know how to ask the question. >> Know ask the question. >> I think that's the key skill, right? because you you need to extract everything possible out of the models and and the skill for that is being able like the Socrates skill, right? Yeah. Ask the right questions. >> Yeah. >> Not spend two days trying to figure out a great SQL query or merging tables and then that's >> and then the question changed. >> So that's all taken care of. Guys, thanks so much. Um you know, if you're out there, you want to get into data, brain surgery is the new model. Understand how the models work. The future is really kind of leveraging the data. We're seeing that now as infrastructure gets commoditized, it's still about mathematics and feeding in those engines with the data is going to be a very big skill. I'm John for with the cube. Thanks for watching.