The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
00:21:38 · Harry Stebbings, George Sivulka
[00:21:38] S1: Okay. And so we are building on Rag to start? [00:21:41] S0: We were one of the first people to productionize the idea of putting a search engine behind an LLM. So you'd ask a question, and then instead of it just replying from its memory, it would actually go and do a search and then reply with that context. So in an enterprise where you have a lot of offline data, we were really the first people to hook up that offline data to large language models to answer that question. [00:22:01] S1: So you're one of the first. And we're we're seeing that now in action, and it's working?
00:22:01 · Harry Stebbings, George Sivulka
[00:22:01] S1: So you're one of the first. And we're we're seeing that now in action, and it's working? [00:22:05] S0: That's a bit of a plot twist over here. We... I actually don't think RAG works at all. It's one of the most used AI architectures in the world, pioneered at Hebia in a very meaningful way. I think every enterprise is experimenting with it, but it has a lot of different failures where a lot of the time the questions that people ask these systems aren't ever explicitly in the data. They're never explicitly stated. They're actually about the data. So for example, you know, if you're asking an AI system, is this company a good investment? Which is actually a very common thing that people ask Hebbio over marketing materials. Maybe it'll say in a pitch deck, yeah, this company is a great investment as a... Something that the CEO says or like a recording, etcetera. But what you actually want from that system isn't to search in the data, It's an answer about the data. Hey. What's the customer concentration? What's the strength of the management team? What are x, y, or z criteria that are are fundamental to our specific investing process? And that's a process. That's not that's not ever explicitly stated. Actually, the marketing materials are often like a load of crap. You have to actually distill what's true out of them. That's what Hebia does. So it's not actually finding something that exists already. It's taking all the things that exist already and starting to answer questions about that information.
00:23:16 · Harry Stebbings, George Sivulka
[00:23:16] S1: Take me to that transition then because we were building on Rack, and we're like, great. We're gonna productionize this, and then we move off. Yeah. And you realize that actually it's bullshit, and it's not as good. Yes. Take me to that realization. [00:23:28] S0: Yeah. So we actually deploy at some of the largest finance firms in the world. We we ended up going from zero to a million dollars of revenue sometime in 2021 or 2022 and raise our series a also from from index, from MICHA index, which is $30,000,000. And we start to see that all these customers... Okay. Now they know what, you know, ChatGPT is. They they know what LLMs are. Hebby has this, like, really mature enterprise product in the market, and we're by far the first to actually get there. And we just looked at all the queries that people were asking. And the questions that people were asking weren't ever, okay. Find me the quote or find me the, you know, command f questions. They were actually more, okay. Read all the documents and then tell me all the times they mention AI or what our exposure to Silicon Valley Bank is during the regional banking crisis. And so all of these questions... It was actually almost 90% of the questions that people were asking these systems weren't answerable by a search through the documents, but rather they had to be work done on top of the documents and then answered.
00:23:28 · George Sivulka, Harry Stebbings
[00:23:28] S0: Yeah. So we actually deploy at some of the largest finance firms in the world. We we ended up going from zero to a million dollars of revenue sometime in 2021 or 2022 and raise our series a also from from index, from MICHA index, which is $30,000,000. And we start to see that all these customers... Okay. Now they know what, you know, ChatGPT is. They they know what LLMs are. Hebby has this, like, really mature enterprise product in the market, and we're by far the first to actually get there. And we just looked at all the queries that people were asking. And the questions that people were asking weren't ever, okay. Find me the quote or find me the, you know, command f questions. They were actually more, okay. Read all the documents and then tell me all the times they mention AI or what our exposure to Silicon Valley Bank is during the regional banking crisis. And so all of these questions... It was actually almost 90% of the questions that people were asking these systems weren't answerable by a search through the documents, but rather they had to be work done on top of the documents and then answered. [00:24:24] S1: My question is, like, what the fuck happens to the rest of the landscape if you're like, no. Rag is not actually the right approach, and they're all loving Rag. Mhmm. Couldn't be hotter right now. [00:24:34] S0: I I don't think they're loving Rag. [00:24:36] S1: You don't think they [00:24:36] S0: are? I don't think they're loving Rag. No. [00:24:38] S1: What makes you say that?
00:24:38 · Harry Stebbings, George Sivulka
[00:24:38] S1: What makes you say that? [00:24:39] S0: I think, like, 90% of enterprise AI right now is almost like this vaporware. We swear it works. Like, look at this amazing demo where we ask, what does the CEO say about the investment? And the minute that they actually go to, you know, try to use it in a real world example completely just fails. And so I actually think that the majority of AI usage, a lot of these usage statistics are all kind of, one of my favorite phrases is fugazi fugazi. And one of things that Hebio really tries to put forth in the market is is to say, hey. Change will take time, but we have a system that that is actually starting to drive real measurable value over very specific defined use cases. And our tagline's always, hey. Stop experimenting with AI, which everyone's experimenting. They're all really excited about it. Start driving value, like, getting value out of it. [00:25:26] S1: To what extent is this RPA versus Ajanti?
00:25:26 · Harry Stebbings, George Sivulka
[00:25:26] S1: To what extent is this RPA versus Ajanti? [00:25:29] S0: I actually am not a big believer in RPA. I think RPA is a is almost not an AI application in the in the new sense of AI. It's like AI in the old ten years ago sense of AI, where RPA is effectively, like, very simple computation. But some of the things that people are asking Hebiah are over 800 page credit agreements or 230 page, you know, SIMS, confidential information memorandums, this marketing material. They're not actually asking for things that... Like copying numbers. They're saying, hey. Tell me what are inconsistencies in this document. Tell me where there's an event of default that we can trigger. There's almost this like open endedness or this new level of computation that people [00:26:10] S1: can do. I think Daniel Dines, who we had on the show, comes out on Wednesday, he said it very well. He said like, listen, RPA is low skilled, low level cognitive processes and agencies, high skilled Yep. Ambiguous Great. Decisions. Yes. I think that's a nice phrasing. [00:26:24] S0: I think it's it's incredibly clear. Yeah. And I think that we very much are capturing the the agent, the the high level ambiguous decision making and trying to trace it all the way down back to individual citations or individual characters that led the model to that decision. [00:26:39] S1: I thought about actually Satya's statement the other day that he made, which is the notion that business apps that exist today will all just collapse into agents. Do you agree with that, and will apps be the predecessor to agents?