Transcript
Hi Mike Matchett with Small World Big Data. We are here talking about of course I it's the thing we need to talk about these days, but how do you get all your data into AI? Ai is really not going to be that useful to you if you're only looking at the the model that they give you and you're not using all your data estate, you've got to connect to it. We've got the professionals here with connections with CData . And we're going to talk about all the good things that CData can do for you and your AI projects coming up. Hold on. Hey, Amit, welcome to our show. Uh, it's great having you here to talk about this. Um, when when you, uh. Let me just get started right into this. When you, uh, looked at, uh, the biggest problems in data integration. The data problems. What got you excited about the connector side of bringing the data together? What was the opportunity you saw in that? I think we saw the fact that the data was fragmented across lots of silos and lots of business application. The trend was these cloud applications were proliferating. Used to be the data would be in databases and data warehouses. But the modern trend is you have data in applications like Salesforce, workday, etc. and we wanted to hit that problem head on and solve it. Right. Because you used to have that one big corporate database with all the things going on in it, whether it's ERP or CRM or both. And you would. You'd run these long running reports and then you would archive that into a, some sort of bi warehouse. And it was all that. But we were a lot more real time these days. And, uh, it's a lot and the data is everywhere. And I know when we start talking about I value the return on AI, you might say it's going to get better for people. The more personalized it is with your own organization's data or with the data that's relevant to the things you're doing. And that's not in the foundational model. So you got to bring your model. Can you just just describe to us some of the things, some of the ways that people can get their data estate into their LLM processing or their generative AI processing. Absolutely. So I think I would like to split this into two big buckets. Right. There's unstructured data like data in documents and PDF files and everything. And there is the industry has come up with a good technique to do that which is rack based model. You you take your data, you embed it into a vector database using embeddings. You can search it and you can provide that to the LLM and provide the right context. And then the world that we live in where there's a lot of structured data, like think about all of the transactions that are happening in Salesforce. And how do you get that in a way that makes sense to the LLM? That wasn't very easy or apparent until very recently, which is what we would talk about. But those are the two buckets I would think about, like the unstructured world and what do you do there, and then the structured data and how you can process that. So with, with with Rag retrieval, augmented generation, um, it's been popular. Lots of people have been focusing on this now for about the last six months or more where, you know, we look at the data lake, the data warehouse, some big archive of documents, and we create these vectors that are similar to the vectors that are used in the LM itself in terms of looking stuff up. And we can include relevant information in our prompting process so that the LM gets all this information and does this. But we're talking about unstructured data and the SaaS data. It's not really apparent to me. You know what how that might work. It's not it's not a Rag process because that data is not already vectorized and sitting somewhere for looking up. It's correct. How does that how does that go? Yeah. So like that's where the industry has come up with this notion that was it was there in the early LMS to where you could register functions. But what has taken the world by storm by is a is a consistent way of doing this called MCP servers. And what MCP servers are is you create these tools where you tell the LM if you need this information, you can come to me with this question and I can answer that to you. So you. So imagine like I need top ten opportunities from my CRM. It can go and query that from CRM and provide. And that might be a tool that can give you this context as the LM needs it. And that's what we were talking about. Um, for the structured world. All right. So this looks a lot I mean, there's a parallel here. This looks a lot like what Cdata has been doing all along with providing basically SQL access to all these data sources. Right? Maybe you could just go step back and say what what your what your emphasis is there at Cdata connectors with this with this SQL process. Yeah. So the good thing about SQL is a declarative language where, uh, LM can very easily generate it. And so we already had that. And what we recently announced for, uh, everyone to try at no cost like this is a technology preview is MCP servers to 100 plus SaaS applications. I think we've launched 50 and we're launching more over time where you could have a you could just chat with your data, you could ask questions Like on Jada. Like, what are the top three tickets or who issued these tickets? And it would come to a response. And the way it's working underneath, for the people who are more technical, it is basically we've exposed tools. Which tools? As simple as here are the entities that are available from this source system. Here's the model of the entity, like the columns and the meanings of the column and everything else that the LM might need to know. And then we tell the LM, like, if you want to ask any question, come to us with a SQL like query and we will give you a response that matches it. Llms are great at generating SQL. They can understand the context of the conversation you are having with it, and they will translate that into a query and and you will get the response you're looking for. I hate to say this sounds like a no brainer when we're talking about intelligence stuff, but you guys had connectors for hundreds of different data sources in this SQL API data integration business that you've had for going on for some time. People would use that for ELT or ETL. They would use this for programmatic access to all these different data sources in real time, and you embedded all the complexity of going out to lots of different data sources with your connectors. And now with MCP, you've just you've really I don't want to I don't want to belittle it and say you've wrapped them with MCP, but you've managed to take all those connected, all that connector technology and bring it into the AI space using MCP. And so it's not this is not really a new thing for you guys. This is just this is just a way to enable it. You're you're very insightful Mike. Like that is exactly right. Like we call these things wrappers in the sense that and I don't want to belittle the work we have already done, but at the same time, the hard work was in figuring out the SQL interaction with the SaaS applications, coming up with the metadata to describe these entities correctly, the columns correctly, making sure all the queries run. That is a lot of work that takes. That took us decades of engineering and testing it out across thousands of customers. CDat a currently has 7000 customers, but that is all wrapped into this new way of consuming data that we think could be very powerful. Right. Right. And just just as an add, I know that among that customer base, there's an awful lot of people that white label what you guys do. It's buried inside the products. So other people could be using CData every day on lots of different fronts and not necessarily have the visibility that that's what's happening, but it's out there. So you guys are using that a lot. Thanks for pointing that out. Like we at some point we like to say like, uh, if you're doing any sort of interaction, you might not know, but data might be flowing through a data connector because our, our products are being used by the most biggest enterprises. You can think of Google, Salesforce, and many that I can't name because they are embedded inside this. Um, but yes, that is absolutely correct. All right. So just to sort of bring this discussion back to the AI opportunity here. So you've got all these you've got all this connective technology. What are the key steps to getting this deployed or in use. Yeah. So there's a lot that's that we're going to build over the next few, um, next few weeks and months. But I'll talk about what you can do just today. Today you could go to our website and download our one of our MCP servers. It is it will work with two of the LMS that have already announced support for it, which is cloud and cursor. So if you're using one of those LMS, you could configure cloud or Cursor to have this ability that we talked about being able to chat with the data. And you can try out the technology as is. And you can do that within ten minutes today. And you could ask the questions that we were talking about and see how that responds. Oh that's awesome. Uh, and if, uh, you know, you don't have cloud or cursor, but you're looking at some of these other LMS, you do have plans to roll those out. I, I assume based on their support for MCP. Correct. So I think, uh, like once, uh, MCP has taken the world by storm, every vendor that we are aware of has plans for supporting it. Openai announced support for it. Uh, Google announced support for it and it in their Google Next conference. Anthropic obviously already supports it, so the notion of connecting in an MCP like fashion already existed. But the neat thing about MCP is like, we could do the same tooling and same infrastructure for every LLM, but we would have to do one at a time and stitch that all together. But with MCP, what happens is everybody is supporting the same protocol. So basically you can have the same MCP server and use it across LLM of your choice. Awesome. Awesome. Yeah, I really think this is a nice second wave after people have explored their rag opportunities. Right? They've looked at their unstructured data sets and they've said, I want to use an LLM to query into my big unstructured data footprint. But now they're going, wait, I've got I've got a significant amount of very specific structured data and SaaS tools everywhere, and I really want to work with that. And this MCP approach would complement that nicely and really complete that, that picture of looking at your entire data footprint. Agreed. Right. I think what Llms unlocked earlier on what people realized. They had so much unstructured data, but no way to query it, no way to get value out of it. And the LLM and the RAC techniques were very instructive in actually unlocking that for people. But the reality is, like a lot of the data is already structured as transactions and these business applications, and we want to make decisions off of those two. And there wasn't a good way of accessing it until now. But now I think you can even get that right. So I mean, if someone wants to to now take the next step practically you've got a website, we can point them out, but is there something specific? Having watched this and being specifically interested in MCP that you would recommend to people, get started. So I would recommend highly trying any of our MCP servers. They're available for free on a free download on the website. We're not charging for it. We want people to experiment with the technology. We call them, specifically a beta. Just because we are, we are improving them. There's a lot we can do to improve the experience, but for people who want to try it out. It's all out there. You could download an MCP server of your choice. You talked about people use hundreds of SaaS servers. Pick the SaaS service that you're using. Most common. Maybe it's asana, maybe it's some Jira, or maybe it's some DevOps tool. Uh, we have uh, MCP server for that and just play with it. All right. And you know, what's interesting to me is when we talk about this, it's not really domain specific or vertical specific, right? Everyone's using LMS inside it, outside of it, every facet of the business and every vertical. So you've got, uh, what, 270 plus connectors to different data sources there? There's something for everyone I guess is what I'm getting at. Yeah, yeah. Yeah, absolutely. Yeah. All right. Well, thank you for being here today. I mean, check it out. You could try MCP for free. Thanks, Mike. It was nice talking to you. All right, take care, folks.
