Transcript
Mike Matchett: Hi, I'm Mike Matchett with Small World Big Data and I am here today talking about the emergence of artificial intelligence, the emergence of AI. Actually, I'm here with emergence AI, and we're going to talk about agents in AI and what some of the cool things are going on in how they're bringing AI technologies to look, data and data science. Uh, but I think there's some bellwether there. You can learn about how authentic AI is going overall, some of the implications for society at large. Stick with us and we'll be right back. Hey, Vivek, welcome to our show. Vivek Haldar, VP - AI Agents: Hey, Mike, great to be with you. Mike Matchett: So there's there's a lot to talk about here. Let's just start obviously, uh, doing there. Uh, how did how did you come to get involved with, uh, AI and application of AI to problems? How sort of backstory a little bit. Vivek Haldar, VP - AI Agents: Yeah. So I lead our AI engineering teams at Emergence. Uh, but before that I had a long stint at Google. Uh, and one of the things I worked on last just before leaving Google was our AI chatbot for support. Um, and this was just after Llms had arrived on the scene. And I was really, really excited by the potential and the new opportunities that Llms in particular, unlocked. So I had been a thing for a long time. You know, companies like Google, many other companies had been building ML and AI systems, but this whole notion of driving them with just natural language was the new exciting thing. Um, and so here at Emergence, we are trying to automate and agent ify a lot of enterprise workflows, particularly workflows that have to do with data analysis, data science, data intelligence. Mike Matchett: So I mean, it's interesting because I see what Emergence AI is doing, and I think it goes beyond data science and data engineering at the point where agents and networks of agents and collaborations of agents start helping us analyze our data. I think it's just a small hop, skip and jump to where we're having the agents start to help us with our business more directly, but that may be getting the cart before the horse. So let's step back and talk about some of the issues with, uh, data engineering that you're helping solve. So when we go to apply AI to data engineering, are we are we are we augmenting a data science team? Are we helping a business user skip data science? What part are you? What part are you? How big? How much of that problem are you biting off? Vivek Haldar, VP - AI Agents: Yeah, a bit of both. So what we've seen at pretty much all our customers, all our enterprise customers, is that they have way more data that they than they can analyze, and way more data than they can actually get actionable insights from. And what we're seeing in enterprise after enterprise is that they have huge amounts of data that simply sit on, analyzed. Now, a lot of these enterprises are well resourced enough to have data science or business intelligence teams, but those teams almost always are also heavily overloaded and are severely bottleneck resource. So the thing we're trying to unlock is to really democratize the analysis and getting actionable insights out of all this valuable data. So the the key insight is that At non-technical folks are able to drive this kind of data analysis purely with natural language and get actionable insights and do the kinds of things that otherwise would have taken a specialist data science or software engineering team. Mike Matchett: Right. So, so you really skipping beyond what a lot of people talk about is low code, sort of DevOps and low code data analysis to really a no code perspective where I need something out of the data. It's data that I haven't really been able to exploit before. And I just create a query in my natural language, like I might do with other gen AI tools. And you're going to go get it for me and process it all with some magic of the magic of the craft. Is that the platform name? Vivek Haldar, VP - AI Agents: Yeah, craft is the name of a platform which allows you to create and save and reuse agents purely with natural language. Kind of one key technical innovation that enables a lot of this is what we call agents creating agents. So at runtime, in addition to using our roster of very capable first party agents for things like data science or web automation, uh, we are also able to generate agents, whether it's through code or crafting other natural language agents that can really solve niche problems that haven't been seen by the system before. Mike Matchett: All right. So I think I'm I think I'm following this. So not only are you taking a natural prompt language prompt about data analysis, building the agents and helping deploy it, orchestrate those. So it's not just here's here's your Python code, but we're actually going to give you the answer back. But in the course of solving some of those queries that the person has, you might need to create dynamic agents like sub agents to go do specific tasks and you're also doing those. So this thing is internally generating more agents as it goes sometimes. Vivek Haldar, VP - AI Agents: Yeah. Exactly. The way the system works internally is that we take the initial natural language query or problem statement from the user, and the system is doing a lot of planning and reasoning to understand that query, and then break it up into a number of sub steps to break it up into a plan. And some of those sub plans or sub steps might have more granular plans within them. So the planning is hierarchical. And then like I was saying at runtime, we might realize that some parts of those plans don't directly map to the agents that are already part of this multi-agent system. And that's when our agents creating agents, uh, technique kicks in. Mike Matchett: Okay. So that's that's kind of crazy to think about because we've all used at this point some gen AI tools, and we've seen some authentic things where you type in a prompt and says, I'm going to do these, these 4 or 5 things, and then it does them, but maybe it gets stuck or fails or comes back and says, I can only get this far. And you're saying in your case, if it identifies that there are some steps it's not already doing or able to do, it creates new agents for that? Vivek Haldar, VP - AI Agents: Exactly. Mike Matchett: All right, all right. That's kind of crazy to think about. Okay. Uh, if I'm deploying, if I'm deploying agents to go this, do I even need a data science team anymore? Is there is there is this sort of a the bell ringing on data science? Vivek Haldar, VP - AI Agents: So I think this definitely enables a lot of people to do this kind of data analysis themselves. A lot of non-technical people, uh, now will have the power to analyze data to get insights from data on their own without having to depend on a data science team. Yes. Mike Matchett: Right. But there's still there's still got to be issues of like security and compliance and things like that that have to be, um, managed or regulated, I think, in an organization. Vivek Haldar, VP - AI Agents: Yeah. Of course, you know, in any enterprise, data security and access control is a very kind of basic requirement. And our system and our agents are completely compliant with an enterprise's existing access control and restrictions. So these agents are only looking at data, of course, that they have permissions to look at that. The user using the agent has permissions to look at. So there's no expansion of permissions. And we're complying with kind of all the the regular role based access control you'd expect in a large enterprise. Mike Matchett: So you've got you've got a way, you've got a way to go after a lot of data. And some of the examples I've seen are really, really I would say cool because that's overused word, but really pretty cool in how you can ask something like, you know, something about the sales data or customer market data, and it will go out and find examples from Reddit, and then go to a corporate database and pull it in and then give me, give me things. And then you can schedule those agents or request and say, hey, come back and tell me that every day sort of, sort of really getting to the point of augmenting, um, augmenting the business and what it's doing. But how can you give us an idea? And I know this is hard to do hand-waving of how hard this is to roll out and implement and, and deploy in an organization. Vivek Haldar, VP - AI Agents: Yeah. So that is another key challenge that Kraft helps with because, you know, often we see that it's one thing to build an agent. So craft definitely helps with building an agent. You know no code local natural language driven manner, but the other half of the challenge then becomes to then deploy and reuse this agent in a repeatable, well packaged manner. And craft helps with that too, because once you've crafted an agent, you can then kind of save it, reuse it, share it across the organization, and it becomes this kind of neatly packaged, reusable piece of AI agent functionality. So we support a couple of different deployment options. We have a fully hosted solution, uh, that users can use hosted on our platform. Uh, and the agents that they craft can be deployed and run on our hosted solution. But we've also worked with several customers where they had more stringent, uh, privacy and compliance requirements, where we have deployed our system into the virtual private cloud or on prem. Mike Matchett: Okay, so there's some flexibility there on how the semantic framework platform actually executes in real time, depending on your, you know, security requirements or compliance requirements. Uh, and I understand that once I have agents in place, you know, it's really it can really help me not only get access to the data, but also implement some of those regulatory compliance issues, PII issues, things that agents might be really good at doing in so helping solve some of the security challenges or compliance challenges that I have. Vivek Haldar, VP - AI Agents: Right, exactly. I think things like compliance and PII are also one of those things that often get overlooked simply because of the quantity of data and the the vast sprawl of this data. And some of the agents we have in our kind of first party roster of agents help with things like compliance, checking against policies with PII, redaction and so on. Mike Matchett: All right. Uh, this this just, you know, just you and I talking doesn't do this justice, right? When, when when you can actually see, uh, and ask an agent to do something and then it does it for you, you feel very powerful and enabled. I have to I have to admit. Right. You know, tell me. Tell me about, you know, the top five products or tell me about who's who's our best customer from last quarter, who's our you know, what are the biggest support issues we have across this thing. And it just goes and does it and gives me that. As a former product manager, I'd be I'm already I'm already salivating at being able to do this for myself rather than having to hire a data science team. Uh, lots of anecdotes, I'm sure we could tell about having to work through data science teams and the queue, get priority, and wait six months to get access to something that maybe is totally incomplete. And now we can just ask the machine in some ways for it's pretty cool. Uh, but if someone wants to actually touch this, kick the tires, maybe get a first hand look at it. Um, where where would you point them at? Vivek Haldar, VP - AI Agents: Yeah. So please check out our website. It is Emergence.ai. You can learn a lot more about the system. Uh, you can request to get on our early preview of craft, or get in touch with our team to get a demo or get in contact with us. Mike Matchett: All right, so a couple couple things. Thanks for that offer a couple things. Just to note summarize what I've heard is that it's not simply about doing no code agents from from natural language query because you guys are nailing that. But you've got this ability for the agents at runtime to create more agents, spawn new agents if they need to, to accomplish specific micro tasks. Um, and then because you have a runtime platform, you're orchestrating the agent network that's in the collaboration that's going on and monitoring it for the customer itself. So it's not simply, here's your agent, go take your Python code now and go run it. It's an entire ecosystem of of, uh, of functionality. So it's all sounds pretty cool. You should check it out. Any any recommendations if you are a data scientist out there. Vivek. What what what should they do? Vivek Haldar, VP - AI Agents: Yeah. I think even if you're a data scientist, you yourself can get a lot of amplification and leverage out of these tools. So check out craft and the the mindset I would encourage is that next time you have a data science problem, uh, try craft, try AI and you might be surprised at how capable it is. Mike Matchett: Okay. And that's Emergence.ai. Vivek Haldar, VP - AI Agents: Yes. Mike Matchett: Check it out, folks. Take care.
