He Raised $6.5M With Three Words

models are at a point where they can do 90% of the work human beings are doing. The biggest bottleneck in any form of AI capabilities is the context that it's dealing with. And if someone has to solve it, it should be us. You can be the world's most intelligent person.

You can be a PhD, but if you don't have access to the right resources, what do you do with all that intelligence? Similarity is not relevance. That line is doing a lot of work. We had like 500 to like a,000 people visiting our website and literally no one signed up.

There was this guy who signed up and he said, "Yeah, and your website is like extremely shitty. You don't communicate what you do. You can have the world's best product, but if you don't know how to communicate your benefits, no one is going to use it. " We just like figured out this one tagline that better databases suck.

" For people who don't know what Hydro DB is, can you fill them in? We are not a memory application. We are the infrastructure that enables memory apps kind of like composio but for a different use case. Hi Nish, thank you so much for being here.

I know you're a very busy guy. What prompted you to say yes to this opportunity? Yeah, so I think one of uh the folks at Compose is an angel investor in our company, Rahul. Um and he kind of helped us like shape up our product the way we want to think about it.

Um, so he's been really helpful since the beginning and then he gave me a call last night's like just asking me to come and do this podcast. So if it was someone else, we would have said no. But uh, yeah, very very glad you're here. So let's get into it.

For people who don't know what Hydro DV is, can you fill them in? So we are the context layer for um, any form of AI application that is being built out there. Uh, models in themselves are very intelligent, right? But models need access to the right context from any form of unstructured business data.

um that could be notion, Gmail, Slack, PDFs, invoices for them to actually be useful in production. So, Hydro DB is basically this bridge between all this unstructured context and your agents in production and we kind of like uh marry the two so that your agents can u you know be productionized very easily. Wow. And what's the biggest misconception people have about what you're building?

That's a that's a very good question. Um so the biggest misconception right now is that people think of us as um a memory layer or a memory application but hydrodb sort of is the infrastructure that people can use to build their own memory layers. Um so we work with a couple of enterprises and we work with some of the fastest growing um AI startups in the valley um that are literally constructing their own memory layers inhouse on top of us. Right.

So I would want to like distinguish between we are not a memory application we are the infrastructure that enables memory apps. Oh, kind of like Composio. Kind of like Composio, but for a different use case. For a different use case.

Okay. For memory. For memory and context. And in general, like so we want to make stateful AI, right?

We want your agents to be stateful from all this unstructured context. And that is what we enabled. Wow. Let's get into more of your background.

You were at Stanford working on a research project trying to build something like Chad BT but connected to all your workspace apps. Walk me through what you were trying to build and where it broke. Sounds good. waited for.

Did it break? I love the assumption. Uh oh, it didn't break. Good.

Okay. Uh so um I think the idea was um for me I think just like a little distinction. So when I was um doing this research project, it was completely for a different use case, right? So we were building search engines but for different modalities.

Um think like uh 3D objects for example, right? Edge case modalities. So that is what I was doing and that work inspired me to start like a different project altogether which was sort of like this um um app that could search all your other apps right so like you have a separate search bar for Gmail separate one for Slack and all of those things so what if you could have like a single unified search bar that search across all your applications and that is what we did we actually were um the number one product of the day week uh and then we were like uh number one or I think number two or three um AI productivity um tool of the year on product hunt. Um so we had like millions of people by the way using that application but I think the reason we stopped pursuing that application was because there's an inherent ceiling with um how like um applica like people want to consume applications right so with with with newer tools like claude and all of these things if you have a very personalized preference of oh I don't like that but I wish it was like that you would rather just go out there and like build it yourself.

So I I do think that interfaces are going to become very fluid in nature. Um so the nature of all your work could be done in for in in the sense like it could be done in your CLI right or within like some other form that we I think I think uh cracking the UI UX for us became a very big challenge because we were getting like oh this is so sucky I was just like the app works why don't you use it right so we were we were kind of stuck in this loop where we weren't able to like figure out what the best experience would be for the customers and then my my whole assumption was I feel like the underlying infrastructure if given to the end user they can definitely figure out the best way to do it themselves and therefore that is where we started focusing with hydrab now. Oh okay. And so with the UX UI like for hydro tob like is it like it's it's very like intuitive.

So we don't have a UIUX now. We only have APIs and we have an SDK. So developers and like non-developers anyone basically can um use our APIs to build their own stateful applications um using our infrastructure. Oh okay.

You had finder. Let's talk about that. Yeah. Tell me more about is finder something you pivoted from the so uh it's very different um so finder was the application that hit like the UIUX ceiling for us right that's the project that allowed you to search across all your applications um using a unified um search bar so that is when we realized that I think what is more valuable for the end user is the underlying ability or inherently being able to unlock all the context from your workplace applications from your PDFs or any articles or tweets you were reading um so that that is what we decided to focus on, right?

The inherent like infrastructure that allows you to unlock context and give it to some form of AI agent so that that AI agent becomes stateful and performs its job in production. So that is where that is why we shifted our focus to just like um using like building hydrob. Okay. So it was a runner up in product hunts 2024 golden kitty awards.

Yeah. You had real traction, real revenue. Yeah. Most founders would have just kept going with that.

What made you look underneath the product and decide a bigger problem was the infrastructure? I think that market or that space in general had a lot of um companies coming up, right? So I think one company I really respect a lot is Glean. So they've been in this space for a really long time and I think Glean is one of your customers or something.

Yes. Shout out. Uh so yeah, I I really respect Glean um for being able to solve this problem early on, right? So I think um and and so Glean is like a prime example of this space and then there were like a hundred other companies trying to do the same problem.

5 million yes but there's uh some interesting developments on that front so we'll keep you updated okay but there's a very Big correction here, right? So, finder was built yes a using vector search. Um, and that is when we started noticing the capabilities or the inherent flaws with vector search, right? Um, vector DBs um for like everyone um um that's going to tune in.

So, vector DBs are basically like this thing that converts text into some numerical representation and then you use some search algorithms to kind of go through those numerical representations to find what is what feels similar um to a given user query, right? U but sometimes that feeling is not enough for you to justify the context that is being retrieved. For example, like the best case example is um if I if I say the word apple, right? The first thing that might come to your mind is the fruit.

But what I was actually referring to was the company Apple, right? And that's like a massive like imagine you were in a board meeting and I asked you, hey, can you give me that um Apple report and you just like brought up these things, right? So, so this is the inherent limitation that we started running into. So we um so what we are trying to do at Hydra DB is we are trying to construct like ontology first indexes which is like um uh supremely and massively different um um from how a vector DB functions.

So a vector DB creates like a flat index of all your documents, right? Um and it basically destroys relationships. It doesn't remember why a piece of context is important. Uh but all enterprise knowledge is n dimensional in nature.

There's like time involved. There's like relationships flowing down and all of those things, right? So we want to preserve those relationships and create an ontology uh first index and then allow agents to kind of traverse all of those like complexities how a human being would and then um yeah that's the difference. Awesome.

Thanks for clarifying. You said you can have the world's best product, but if you don't know how to communicate your benefits, no one is going to use it. Yeah. Walk me through what that actually looks like.

And yeah, this one's uh kind of personal. So, so what happened was we uh so we started like working on Hydra DB and like yeah before we had like this um the messaging uh I I don't even like like um precisely recall what the messaging was but it was something um extremely technically irrelevant. We thought we were the coolest company on the blog to have figured out, you know, uh this like new shiny messaging, but like in reality, no one cared, right? Literally, no one cared.

What was the shiny messaging? Like how I think it was um h I think it was something around like build scalable lightning fast AI or something like that. Something along those lines, right? And literally I can assure you no one cared.

Like no one freaking cared. " And he said, "Yeah, and your website is like extremely shitty. " And that's when I realized like you could genuinely have the world's best product, but if people don't know what you do, like no one's like no one has enough time to figure it out on your behalf, right? So yeah, that's how we Yeah.

Wow, that is such good advice. So it's like kind of like moving away from like the shiny like flowery language and more so just like straight up like this is what you do. So I mean I think for example like even with our recent launch right our thing our our main goal was to generate a little bit of awareness around like what are the fallacies that we ran into right or what are the challenges rather we ran into um and the fallacies of like vector databases and all of these things. What we did was at the end of the day we just like figured out this one tagline that vector databases suck and we were like upfront about it right and then people were like oh you are right right and we got like 200 companies to sign up for demos and like we we we um decided to like get on a call with these folks and we were like what is your problem can you be honest with us like vector databases actually suck can you help us replace them right and we were like sure let's work together and that was like the new like that's like the tagline pretty much wow so like same product different wordsusto customers started converting from competitors.

How long did that take you to figure out? What was the moment? Oh, yeah. You already said that.

When was the moment you knew the messaging was working? Yeah. Sweet. Amazing.

So, most technical founders hate thinking about messaging. You clearly don't. Where did that instinct come from? I think, by the way, we have a great team.

Um, I know Karthik's off camera, but head of grown. Yeah. Uh, so he's been uh Yeah. So, I think I think it takes like a team of people to do it.

I don't think it's like so if you are a founder you have a very particular like you have opinions that are very strongly held in your belief that oh this is the right thing that you need to do right um I I do think it takes like a team of people um who can just like spread out there and figure out different variations and combinations like like it's literally like how many permutations and combinations can you try before something works um so that is exactly what we did were you doing growth engineering um you were doing growth yeah uh product and growth product product and growth so maybe as Well, they probably I don't I think um like from a the growth is very different or like just like making sure that it's Yeah. you going from like a million visitors to like 10 million visitors is a completely different things. But if you have zero visitors that is different. So that's completely different.

Wow. So you really had to and finally after a lot of iterations like you found what worked. Yeah. Cool.

7 million views on X. Yeah, probably more now. 5 million to kill a vector Dav business. Walk me through how you wrote that tweet.

Was it a one draft or like 20? Did you like um did like team help? Yeah, I think it was like um some n number of attempts to write that draft. It definitely it wasn't the first draft that we came up with.

Okay. And then uh it was like even paying attention to the like even the smallest of details, right? That was the goal. So for example, uh we want to say that vector databases suck.

Should it be the first thing that we put in the tweet or should it be the last thing? Um what's the right balance between our claim of saying that vector databases suck and why do we want to say it? Right? So tomorrow you wake up and you have this like random person on the internet shouting that oh vector databases suck.

Why would you believe that person? Right? So it had to be like we had to figure out oh if we make this claim what is the justification behind this claim that customers would actually understand. Right?

So it was like a bunch of like iterations that we had to do to find the fine line difference between oh this is where we want to just like talk about the problem but this is how we want to explain why this problem matters. So it was like a so why the problem matters is so important like for us it's very important right because I think uh clearly um there we have spoken to enterprises that have been trying to productionize I'm not even talking about improving I'm trying I'm saying productionize AI for the last one and a half years. What do you mean by protected? Like uh they don't have a single AI agent running like for a customer um support function or any function as a matter of fact, right?

And like we get on a call with these um folks and they're like, "Oh, we've been trying to build something in AI but nothing works for us, right? And uh the specific uh numbers we get is like, oh, we have like x documents or like x things that we want to uh want our agent to, you know, just utilize, but we're unable to do it, right? because we don't get the accuracies that we need or we are unable to just like um ensure that our our agent retrieves the right context and we need to meet certain standards like because this is an enterprise right this is not a startup that can be like lazy with their um um accuracies and all of those things so so because they have to meet certain standards and because nothing has been working for them they haven't been able to build a single AI application in the in the last one and a half years right and that's kind of like frightening um so so we just wanted to ensure that like enterprises who saw our launch could understand why this was happening with them. Right.

So this is where we wanted to find some balance. Okay. Okay. The line similar.

Sure. Relevant. Similarity is not relevance. Yeah.

I think you're talking about that. Yes. Yeah. So I I think that line has like a very like um solid impact, right?

Similarity is not relevance. Um most search systems are working on that principle right now. But going back to our Apple example, Apple two different things. Wow.

Wow. So just because it's similar doesn't mean it's relevant. Exactly. Right.

There could be like a million Julia in the world but like that Yeah. The name being similar doesn't mean that they're the two same things. Right. Wow.

Yeah. And that's like that line is doing a lot of work and I mean it's in a very few amount of words. How do you think about making a technical concept that specific and that punchy? H how did I think about it?

That's a very interesting question to be honest. I think just my brain like like I I don't think my brain has the right answer right now but um I guess when you have enough conversations with the people around you um and then I think it's once again it's an opinion that's shaped up like after like any number of conversations right um to be able to like again I do think that this tagline is very important for us that similarity is not relevance right just it's it's four words but that could literally translate or just make you understand or make you shift your entire entire business from using a vector DB to something else, right? Um, and that could literally transform your business entirely by making sure that you can now build something with AI, right? And this line is so like impactful that we I think we came up with it over like a I don't know.

I think it took some time for us to figure out like this is the best line that we could use to convey the feeling. But I think it took a lot of conversations and talking to customers and figuring out what the best way is to condense this. That is awesome. 5,715 bookmarks.

You got that many bookmarks? We expected more, but okay. You expected more? Yeah.

Okay. Well, that means people saved it to come back to it. Mhm. What do you think they're saving it for?

And why did you expect more? Yeah. I mean, um, for us, the ceiling is like the sky is the You always make sure to like have like a bigger goal, so it's like always a higher number. No, it's because we'll achieve any goal we set for ourselves.

Yeah. Love. Yeah. But the thing is I think I think most people might again I cannot speak to why people bookmarked it but like if I was seeing my own video the number one reason I would have bookmarked it um is because this would be the first time someone to I mean principally speaking in the last 2 to 3 years almost all AI applications have been built on top of vector databases right so even if you log on to like chat GPT and ask a question like oh how do I build my own AI native email calendar right or email and calendar application Um the first recommendation that you get is um oh you need some connecting capabilities or something like data connectors and all of those things.

You need to figure out O um and then you need to like extract all the context you then you need to store it in a vector DB and then you need to make calls to that vector DB and just like extract the relevant context that is needed. So for the principally speaking like in the last 2 years most people have built applications on top of vector DBs. But like I do think and I very strongly believe in this that most organizations and enterprises that have adopted um like this notion of vector DBs have started experiencing like a ceiling um like a ceiling with the capabilities that their AI applications can literally do right um and I think most people were unaware or like maybe they just were like I don't know maybe they just like yeah let's just say they were unaware of why this was happening um and I I do think like if I was watching my own video I would have been surprised right I've been running my application with the last two years on a vector DB. I know it's not there yet, but I don't know why.

And I see this video and it's like, oh, because vector databases suck, huh, that's interesting. Why don't I just like save it, revisit it, and see or maybe share it with my team later on when it's relevant. So, maybe I think that's the number one reason. Nice.

Yeah, I love how you're like you're very you're very like reflective on that and like you're very like kind of Yeah. So, vector search returns similar results, not relevant ones. Give me the most concrete example of what the failure looked like in the real world. The moment where similar is not good enough.

I think we've been talking about the Apple example. So yeah, the Apple that's a good example. Okay. 95% of Gen AI startups fail their pilots with enterprises because their AI can digest context.

You called you cited that MI that's like an MIT stat you cited. What are those teams getting wrong and what would they have to change? Yeah, by the way, I cited it just because it was like a fun thing to site. It's not like a scientifically proven thing that the context is like the main reason but I do think like a majority of AI applications are unable to go from like PC's to production because I mean even when we were talking to teams and like some of these companies were like series B companies right and that that are building like some new agents for um their end customers I mean it was genuinely shocking like for them PC meant running an entire um like like literally uploading an entire PDF into a model like claw or ggemini right?

And then that's a P for them. But a productionized system actually looks like having a data set of millions of those PDFs and you obviously cannot digest all this context in a single LLM call, right? And I do think that's a very big um gap between like how PC's are being built. You take one model which has like some which has some decent amount of like context window.

Um you upload as much context into that model as possible. You show a demo, you show a flashy PC. uh but when it like when the time is just to right to convert into a productionized system you are dealing with a completely mass like a different a massive corpus right of knowledge and then how do you refine search and retrieval and um just making sure that you can scan these millions of PDFs in production and I do think that's like a very major gap between how PC's and production systems have been building. So that's why I just wanted to have fun with that.

How does hydrob solve retrieval differently? Not like the pitch version, but the version you'll explain to a technical founder at 11 PM who's frustrated. I mean, so if it's a technical person, we just tell them that we build ontology first um um indexes and we process all of the data objects in memory. " Nice.

So you mentioned at the beginning that uh Raul connected you to the Composio ecosystem because Hydrob and Composio are solving adjacent problems. How do you think about the relationship between the integration layer and the retrieval layer? Yeah, I think um like the first half of the equation is definitely setting up some strong integration capabilities and we have done this thing right where where some of our customers that were looking for some integration capabilities um we looped in uh teams like team members from composure and now we have like shared channels between like our customer X us x composio right so that's still happening so I I do think that we want to make our customers successful like if you come to us and you tell me that oh I'm building like a deep research agent, but it has to take things from my notion, right? Um like the way we want to help you is we want to make sure that all your integrations are sorted out.

We are not an integrations company at the end of the day, right? Um neither do we want to like that's a completely different problem set. Um so we what we've done is we just like call up composure and we tell them listen like channel uh let's you figure this out and then the second half of this equation is actually the context management context retrieval and that is what we do. So I I do think like both things go hand in hand.

So, Composio handles connecting agents to tools. Hydrob handles making sure the right context comes back from those tools. For a founder building an AI agent today, what does that full stack look like? Hm, that's interesting.

Um, I would break it down into like three or four major verticals. I think uh picking the right um observability stack is something that I would obviously start with. I think most founders or most builders I talked to I they were like of the opinion that they don't need observability and traceability into their systems right but um that changes very quickly as soon as you have like 100 or thousand or like maybe like 10,000 agents in production you need to know what's happening under the hood right and that's why like you had earlier you had analytics for simple products like SAS products and now you have observability and traceability for these things um I think that's one thing um the second thing is having the I think most AI that is going to be built moving forward has to be connected to your life, right? And your life is your work and personal stuff and all of those um applications, right?

And this is where something you use like a composure to make sure that you can have like access to all these tools and data connections and and then the third thing is of course like a context management layer um that you have and uh making sure that you like store all of these context man like pieces of context retrieve them and then that is what makes your AI stateful. Um I think the final layer is that I see a lot of founders now like think about is um having like a um maybe a smaller model but having something fine-tuned to their own use case right um I do think like SLMs for most use cases are like much much more powerful than throwing like the most powerful model like a claw at every single problem. So I do think that there's like platforms that allow you to host your own SLMs fine-tune and train them. Um so I think that's like the fourth quadrant that that completes the entire uh picture.

Cool. You said you've been connecting Raul with founders who need composial like capabilities. What's the pattern you keep seeing? What problem are those founders running into?

The number one problem is that um listen, I have this idea in my mind that I want to get out there in the real world. What's the shortest amount of time in which I can get it like from ideation to production right and um the the the pattern is simple. most of the applications and that's why I was also indexing on this statement that most of the applications moving forward will have a component of um data flowing from these applications rather than some generic PDFs being uploaded right I think we've moved way way ahead of those like AI chat with PDF kind of applications right um so if you're building an email assistant you need access to Gmail if you're building like a like a designer agent then you need access to Figma and Slack u product is like linear and curs and all of those things so I think like a holistic like connection of ecosystem that's happening uh for most of the people that I now talk to. Um and I do see like most of the use cases that are being built on top of Hydra DB are literally pulling from applications most of them um instead of like some generic data store that has been there you know with them.

If you were starting an AI agent company today what infrastructure would you buy and what would you build yourself? I mean of course I would start with Hydrav. Yes. Um so it depends on the use case by the way right so if you are building like a voice um AI support agent then I would look at like um some some foundational voice companies um that are providing you with models um if I am building like a food delivery agent then I would of course look at um companies that allow um the agents to make purchases on the behalf of users and there's companies that allow you to like manage inboxes for your agents autonomously and all of those things.

So it's it's a very like I think it's a very u wide spectrum of tool choices that I would have to make um for me to decide what sort of AI agent I'm building is the most important thing I would index on and then just like keep working backwards from there. Awesome tip. You go by contact king CEO on X. I love that username.

That's definitely not a modest handle. Yeah. What's the conviction behind it? Originally I think um yeah it was just my name.

uh but then we like um did this because I the conviction is like it's it's it's not about the Twitter handle. I think the conviction is in the broader problem statement of ensuring that like literally I think models are at a point where they can do 90% of the work human beings are doing right literally speaking like like researching about someone's background or maybe writing code and doing all of those things right I do think the biggest bottleneck in any form of AI capabilities is the context that it's dealing with right um and if someone has to solve it it should be us and therefore we want to be the forefront like the the the frontier the polebearers of this um like space. Yeah. Because like AI needs like you need to someone needs to control AI like Yeah.

I mean it's the same as you can be the world's most intelligent person. You can be a PhD but if you don't have access to the right resources what do you do with all that intelligence? Right. Yes.

Exactly. Mhm. How was your journey building in San Francisco? Yeah.

How was your Yeah. How was your journey? How was your journey? Yeah.

still going. It's Yeah. Is it amazing? How has your background shaped you?

How you think about building for a global market? I think um Okay, so SF is anyone that's um willing to like take a bet, I think they should definitely move to SF. And it's like very classic repetitive advice, right, that you hear on the internet. I'm seeing like 17year-olds on Twitter like moving, dropping out of school and moving desktop.

That's so interesting. I I know, right? It's just like when I was 17, I was like shuffling Pokémon cards, but like these guys are moving to SF, so that's very interesting. But yeah, it's classic repetitive advice.

If you're if you're willing to like shape the future, um then you have to be in SF because there's a lot of like new things that you are exposed to on a daily basis like forget company building, right? the the spectrum of things in terms of like what we can do with even Hydra DB was um like 10x or like 100x amplified with respect to when we were talking to like researchers working at frontier labs and they told us oh you should be thinking about this problem statement because this is how models are being built literally and the models are being built here right um so so I think just the rate of learning exponentially becomes higher you get to hear words that you might not have even have heard of on the internet right um and I I I remember like debating like the Stanford PhD on like this this thing about like test time training and he was convinced that uh this is going to change the context space and all of those things and we I would have not had that discussion anywhere right anywhere else so so that's the number one thing I think being in SF gave us like an edge over some of the other companies that are building in the same sense at least trying to and then the second half of this equa uh equation is like um yeah I I I do think that I already had like a few friends that were founders um so and I wor and I was like very lucky to have worked with them um and they were based out of the area. So I the the one of the re like um first times I came here um was like this friend was building like this uh AI meeting recorder of sorts, right? " And that was the in like I worked with the founder like he was the founder and then um like I really loved his energy, right?

And then that was the first time I was exposed to um sort of like how founders operate in the Bay Area. And the way they're thinking about like you know world domination all the time is so different compared to anywhere else in the world. And that's why I think for me moving here was like the best thing we could do to just like move the company here. Wow.

Yeah. Cuz their energy is like they're very like like world domination. It's not just like dominating like a niche or like a section of it's like world domination. So interesting and it's while it's aggressive in its approach but I do think like some of the most fundamental ideas like come from that sort of aggression.

Do you see yourself like living here like for a while or yeah like even when you're when Hydro DB like scales really like really big is living wherever our customers require me to live. Got it. So this okay wherever because I do know a company that end up moving to New York because they realized exactly that customers so move where your customers are. San Francisco is amazing, but if your customers are not there, maybe it's probably or maybe you've just saturated the market, right?

What if you're working with every single while theoretically it's not true, but like what if you started working with every single company or every major company that you wanted to work with in SF? Then you might have to move base again and again and again till you get to a point where you get to choose where you want to be, right? So true. Point.

Amazing. What are your values personally? Like not the companies like how like what do you actually operate by? That is a very interesting question.

My personal values, huh? I think I kind of optimize for um like learning from people who I know have done something that I really genuinely want to learn about. Right. So like um I have I have Yeah.

And I've been lucky to be in like the same rooms with these folks. So how do you think of um so for example like one of my notions that were recently built up were um around like how fluid interfaces would become right and this was like I this was a conversation with Rahul by the way right and I was talking to him about like oh what do you think of the future of interfaces like my personal belief was um like interfaces would become so fluid in nature that you could claw code your own interface into existence but he is of the opinion that um interfaces will cease to exist overall and that makes so much sense right like I'm yeah it's just it's like Why would you need an interface if like a single CLI command or like a text message could get the work done? Like most applications out there are there to help you navigate some complexity that otherwise would be very difficult to do like your banking apps, your your voice and chat apps and all of those things, right? So so I think these are different notions that I kind of I I am trying to adopt and just like understand.

Um, so this is one thing I really genuinely value a lot that I can um like I I love to hear different opinions um from people um who have specifically like there needs to be some weight, right, as to why I'm listening to your opinion. So this is like a twoprong thing. Um and that's the number one thing I value. Yeah.

Yeah. Cuz like you like taking opinions from just anyone is like this is like the number one thing I've learned in SF. You will meet like a thousand people who have like a thousand different opinions. But if you listen to all those thousand opinions, you're doomed.

Yes. You have to know like the person the background of them and who Yeah. Do you feel like you're very like you're you can you're very easily like you can easily like do you easily trust people or you easily like to like No, I'm a deeply mistrusting person. Okay.

Unless someone careful with Yeah. I mean when you're for example like one of our um angels and I of course like he's a superhero for me like Jeff Dean right he basically practically invented like Google search and all of these different things right um when you're in a conversation with him then you start realizing what the value of a real like expert in the field looks like versus like an unformed halfbaked opinion looks like and you're able to distinguish between the two um so I think um that that is really helpful what's the distinction if let's say the person who invented Did Google search practically turn? Oh yeah, that's Yeah, that is like the big Yeah. Who are the people who shaped how you think?

Mentors, founders, thinkers you keep coming back to. I think the number one person who's had like a very uh big impact on my life is and I hope he watches this. I'll actually send this video to him afterwards. Yeah.

His name is Peter and he's the guy who actually I I texted to to see if I could work with him and come here. Um so like he was running this company called Ber. um and I worked with him for a while. Um so yeah, he's the number one person like I still give a lot of credit to because he exposed me to these things.

Um the other person I really value is like um the founder of smallest AI named Sadashan and I did like those walks with him. He helped me shape a lot like he literally like helped us um shape how we should think of AI agents and the future of context management all of those things. So those are two people I really um yeah value. That's so amazing.

5 million. Mhm. Walk me through what was the process like? What did the process actually feel like from pitch to close?

We by the way um um closed our entire round in like 2 weeks. So for us it was very fast, right? And um I think it was easy for me to do because we already had some pre-established traction. Uh we had a product that actually worked and um I saw like a couple of investors just like wanting access to our APIs and the reaction was holy it actually works.

Right. Wow. That's amazing. Yeah.

So that was the kind of reaction and then therefore we were able to close the entire thing in two weeks. Um so I I I do think like having Yeah, just I think I'll just stop there. Okay. Great job.

Wow. Was it like harder before? Like this is like easier like I mean when you obviously have a little bit of like a runway then you can take some long-term bets right before you have runway. It's like you have to optimize for the smallest of things.

Um so we had raised like a small preede round of like 400k before this and like we had to like like building something which acts as the infrastructure for millions of companies out there is a hard problem to solve right um you need to take care of latencies accuracy making sure the thing scales making sure your docs are good and there's so much that goes into it and then optimizing all of these things with like a um like a smaller runway is a harder problem to solve than like having a larger runway and then taking massive bets. Okay. What's keeping you up at night right now for hydro tob for like the industry for you? I definitely um so there's like a few things that are definitely like fundamentally LLMs and I have spoken to like my friends who have been working on this but fundamentally there's like a massive shift in how some of these labs are thinking of building future LLMs right so one of the things that I've been um deeply thinking about is I do think that the future of um models is split into two halves right where you have like very large uh models running deep inside your data centers um and then you have like smaller faster models running on your mobile phones running on the edge.

Um so and but but the one thing that remains common in across all these things is they would always need the right context to function right. So my hypothesis is like there needs to be like this thing this infrastructure that helps like these larger models that are solving very complex problems like rocket science like nuclear medicine and all of these things. um to to to how do you build something that gives context to these things versus like your average Siri like application, right? Oh, can you call my mom?

Can you send my shopping list to this person? So on and so forth. And the SLMs that are running on the edge, they need like smaller context stores which are running on your device, right? So how do you build something that is running on your device at the edge without talking to or communicating with a server?

So this is like a wide spectrum of problems that I'm thinking about and how we should move forward with it. But uh this is very yeah this is this is very interesting if you think about it. Wow. Thanks for sharing.

What's the version of Hydro DB you're building towards? The one that it works changes how AI agents retrieve contexts forever. Again I think this ties back into like the previous problem statement I've been thinking about. I think a future version of Hydro DB which is very relevant or like one of the variations that is very relevant is something that runs locally on your mobile, right?

It's so small, fast, intelligent and efficient that all your messages, your personal information, all your photos, it has access to that context. So that you have like this this centrally running AI that's controlling your phone. Your phone knows everything about your daily life, but you are safe and secure in knowing that none of my data is actually leaving my phone, right? So what does that version of Hydro DB looks like is something that we've been thinking about.

Uh but I think once we make that a possibility like there there's like a billion devices on the planet. Um there's millions of phones, millions of laptops and if we could bring context locally to these um um devices, I think we might be at the edge of solving a real uh problem for most AI companies out there. What's one thing you know now that you wish you'd known when you were first starting as a founder? Okay.

H how long have you been a founder? For how many years now? Two years. two and a half years.

But I think okay, between now and when I was starting Hyderabb. So there's like a bunch of things. I'm trying to narrow down to like one specific thing if I could. Um but um h I think one thing Yeah.

I Okay. Okay. I think this is like basics 101 is um I think the right messaging attracts the right people to use your product and that is what a right product looks like. Yeah, you're a really good example of that like that whole transformation.

such such great advice and yeah, thank you so much the whole podcast. I really appreciate you. Awesome.