Bob van Luijt: 0:00

I call that the magic of machine learning. Byplan JavaScript. Let us help you explain how we do it rather than show off. And I just had this idea that, hey, wait a second.

Dr Matthew Mirman: 0:23

Welcome to the accelerometer. I'm Dr Matthew Mi rman, founder and CEO of Anarchy. Today, I'm delighted to be filming in Amsterdam, where I'm being joined by Bob van Loet, the CEO of the most legendary startups of this age of generative AI Weevee. Weevee is a vector database which fundamentally exists to enable deeply semantic searches, image-based search, aaa-enabled recommendations and more. These kinds of databases are not exactly new, but Weevee took the field a step further by being truly first class about embeddings and being open source. As Anarchy is committed to open source infrastructure, I'm really honored to have Bob here. Can you tell us a little bit more about Weevee? Yeah, sure so.

Bob van Luijt: 1:08

Weaviate is founded in, or like, actually not the companies. We have the company Weevee, that we have the open source project Weevee. So the project was opened and I think that was in 2015. And this has you know, embeddings played an important role in that, because I was working with embeddings but it was not a vector database yet. So what I was working on was I was working on a project where I wanted to improve search results and early on, that happened with academic texts and not long after that it was related to IoT projects and what I wanted to do was this I wanted to solve the problem that there were, like, different manufacturers and they had certain types of documentation or language in the APIs and what was coming out. But the problem was was that these different organizations had different language to describe similar products. And back then I was working as a freelancer and I joined a call.

Bob van Luijt: 2:18

I vividly remember this guy named Paul Grof he was actually on our own podcast, I guess, and that was the first time that I was introduced to vector embeddings and this was like Glof, and so single word, embeddings. That's you know where it all got started, and I just had this idea like, hey, wait a second, what if I don't try to make certain mappings or certain graph structures, but what if I try to use the centroids of the words in the embeddings so for people listening who don't know what that is? That's the vector. Embedding is a dimensional representation of the word and if you have a sentence of words, you can group them together, pick the middle and then represent, in this case, the data object. Am I able to retrieve the data object by searching true vector space? And that worked okay-ish. It was like, I mean, it's not where it is today, but it was working okay-ish. And I was like, hey, there's something here.

Bob van Luijt: 3:20

And maybe a year later or something, I went to Google IO. I was sitting somewhere in the audience and I remember that Son of PJ he said, actually, fact check this. This is actually true. This is that. He said we're gonna move from mobile first to AI first. And I was like, oh, I know what you're doing. They're using vector embeddings to index these pages or whatever the ads or whatever you and I was like this is a thing. And then I met my co-founder, hn, who's currently the CTO at VV8. And he was like, hey, wait a second. There's actually a huge opportunity from an architectural perspective, to build a purpose-built database for vector embeddings. And if you go back in history you see that the first iteration of VV8, we call it a knowledge graph and they were like ah, then we had the word smart graph and then it turned into vector database and it always had the vector embeddings at its score. But I just where vector embeddings are now and how everybody is using them. I could not have predicted that, like almost now and nine years ago.

Dr Matthew Mirman: 4:39

So back then, where people bring these embeddings with like word to back yeah, I remember being at Google when that was happening they were really pushing us to use word to back in all of our 20% time projects.

Bob van Luijt: 4:53

Yeah. So what's so interesting about it is that it's a. I call that the magic of machine learning, and what I mean with that is there's something and I never found a better term than, as I like to call it magical is that the moment that you move through vector space, but the way that you move through it by adding words, subtracting words, those kind of things, and something logical comes out from a linguistic perspective, that's magical. And I remember also when I was on stage presenting, presenting how we've had worked like the early. So the API today is not much different from back in the day. The whole database itself is, but the API isn't. I mean, change a bit, but not a lot.

Bob van Luijt: 5:42

And that was on stage and people ask questions like so you have like keyword napping somewhere? I was no, no, no, no, no, no. So it took me a long time with the people Google and other startup founders to really explain to people no, no, no, this is a completely different way of representing data and it's just yeah. Sometimes I have nostalgia back for these, for those days, because it was so much fun traveling around and showing people what you could do with the technology. So it's yeah, so that yeah.

Dr Matthew Mirman: 6:21

What was the first aha moment that you got when you showed it to somebody?

Bob van Luijt: 6:27

You mean my aha?

Dr Matthew Mirman: 6:29

moment, or for the For somebody else's aha moment.

Bob van Luijt: 6:34

I quickly figured out that if people tried it out themselves. So I was on stage and people were like, huh yeah, but how does that work? And at some point somebody and I forgot where it was, but somebody in the audience, because the first data set that we had were news articles and there was somebody in the audience who, just you know, yelled a query so how does this query perform? And they just yell at us like sure. So I was like tap, tap, tap, tap, tap, tap and write results, right. And so I was like I need to make this a thing. So I showed people the data set. I said, like you know, I have, like from the New York Times, there's a handful of articles, right, just the Atlantic, whatever you just. And I was just asking me something, asking me a query, and you know, and I'll just type it into the API and see what comes out. And that was really the aha moments for people that they were like, oh, there's no trickery here, there's no, this actually works.

Dr Matthew Mirman: 7:33

So what is some idea or a thought that you have that you really want to tell the world about before you die?

Bob van Luijt: 7:46

I'm now focusing a lot of on work, so it's something work related, because I just, I just it's something, and then I'm starting to. I was when I started the business. I was convinced that doing my job well was doing as much as I could in a day, right? So you know these tweets, right, or how do we call them nowadays accesses and where people go like, oh, I get out about five AM and then I do blah, blah, blah. And I got convinced that the exact opposite is true, that if you run any business, the best way of doing it is by if you're able to do absolutely nothing, and I'm happy to double click on what I mean with that, but that is what I'm convinced about. So success is being able to run a successful business by doing nothing. I like that.

Dr Matthew Mirman: 8:42

I have a reminder. You have this book that I've already called who, not how, and it was just like anytime you're thinking about a business, anytime you're thinking about like whether you want to be doing something or not, it means that you're not good at it and you should find somebody else to do it.

Bob van Luijt: 8:55

Exactly. And so one of the things that so I am guilty of something and that is a, and I'm so I really want to write this down in blog posts and when I just you know when I really got to the core, the essence of this idea but I'm guilty of a Google search. When I started Weavey8, but Larry said, what does a startup CEO actually do? I was like I don't know what to do. I mean, I was super busy because I was like Defrel, marketing, product web designer, the whole shebang in one. I was just doing everything right. And then when the company and then one of my co-founders, the channel, was like working on the core database, right, but everything around there, I was kind of working on that. So then when the company grew, I kind of figured out there's three things that I need to do, and I need to in no particular order. I need to make sure that there's a great culture, that people like to work at the company, or that we find the right people to join that culture. The second thing is to have a vision, and vision is a very complex thing because it's better to have to articulate a vision rather than keep searching for one. And the third one is just having on money, right, and money in the most ideal way of getting money is, I would say, is revenue, and but if we're still building, then maybe there are investors who want to help us, right? So that's it.

Bob van Luijt: 10:21

And how I visualize that is you have these spinning disks, so on a stick, right? So why you need to spin these disks? And those are these plates, I should say. And it's basically you have these three plates, they're spinning, and the better you get at spinning these plates, the longer they keep spinning, right, and if all three of them are spinning in harmony, you can just sit down and relax and let the people in the company do what they're good at. And then, if something starts to wobble, and oh, you spin it again, right. And that is where that idea comes from, like if the utopia, the best way of running a business, is if these or these plates keep spinning these three plates. So that's where the idea comes from of saying, like the state of utopia in running a business is doing nothing.

Dr Matthew Mirman: 11:14

I like this analogy of spinning plates, very similar to juggling, but just one of these things that very few people can do. Yes, Not everybody can be like completely blunt, not everybody can be a see.

Bob van Luijt: 11:26

No, exactly. And the thing is that the and I also don't think that people, that everybody should want that, because it's just a. I mean, I sometimes go to these events where, like you, have all these CEOs and this is weird bunch of people. So it's like, you know, it's a. I mean it's fun, it's like you can definitely have a laugh, but it's like a. It's so interesting and it's so diverse.

Bob van Luijt: 11:56

But the point is that you somehow and the marketplace role in that right. So there's a. If you have a market fit, there's a market pull and then you need to figure all these things out. So you're super busy because you're constantly tweaking and optimizing the plate to, you know, to spin properly. It's like because if you make a mistake, if you sit down too soon and the plate falls off, your company dies. So I mean it's tricky, but it's a.

Bob van Luijt: 12:23

And also, one thing that we do in the company is that we, so our organization is very flat, so we're now like 60-something people and there's only one step between me and the rest of the organization. And I did that on purpose, because I always like to say it's like, yes, there's these tiny hierarchy at Winfrey, that's like in this one step in between. But if you want to see expertise, you have to flip the chart around, because the expertise in all the people working and putting their shoulders to the will to make things work and I want to keep it as long as I can. I want to keep it like that because I think that's the most efficient way, the most organic way to keep base and keep moving forward.

Dr Matthew Mirman: 13:14

This is going to ask us like I was calculating it, like I had that's like three one-on-ones per day at least.

Bob van Luijt: 13:21

Oh, no, no, no, no. So this is I mean sometimes, but the thing is the. That's another principle. It's very trust-based. So one thing that I do that people listening that might be thinking over on the start that might find is interesting.

Bob van Luijt: 13:39

So a question that I sometimes get is like, well, if you have so many direct reports, how do you manage that? And I manage it by sampling. So there might be our developer growth team might do something and they organize an event. And there was recently was an event in Amsterdam and I was in Amsterdam, so like we are now. So I joined that event and if I would have been in New York, I would have gone to the New York event, whatever right. So I went to the event and it was fantastic. Everything was great, it was well organized and it was so many people, it was just great.

Bob van Luijt: 14:18

So, based on that sample, I would be like okay, so the events that the developer growth team is organizing are great. And or, if I want to try something out, I read a little bit of documentation and if the documentation is great, then I'm going to assume that the rest of the documentation is great as well, and that's how I do it. So I only interact with people sorry as in me towards them. Otherwise, I mean if they ask me a question. Obviously there's also another path of interaction. When I'm concerned about one of these things, that is like hey, what's going on here, or those kind of things. So it's super organic.

Dr Matthew Mirman: 15:02

So back to this idea of spinning plates, I actually have a fun anecdote that I really wanted to share. So when I finished YC, I was going through a co-founder breakup and I was just super overwhelmed by everything that was going on. Like I was like how do I actually run a startup entirely alone? I gotta go and I gotta speak to customers. I gotta be building this product myself. I think if I go and talk to investors myself, I gotta go and figure out marketing and they have to be going and hiring right now.

Dr Matthew Mirman: 15:31

Like, yeah, I went to my partners like, how do I do all of this? And it's just like you know, actually you're running it. You can choose to do one of these things at a time. So I did that for a few months. I came back a few months and I was like, okay, well, I did all those things at a time. I feel like I'm falling behind on each one of these things now and they were like well, actually as a founder, you have to do all of them at the same time. Again, yeah, I think at that point, like this notion of juggling really finally clicked to the spinning plates. And yeah, I guess I'm just wondering, at what point did all of that come together for you?

Bob van Luijt: 16:09

So when we did our series A, I was able to hire a couple of people to get this. It's called a cell structure. It's by I'm blanking on his name, it's a Dutch entrepreneur, Winston is his last name, but I forgot his first name. But is that cell structure where you basically say, okay, you have a cell structure, you have a cell and it's like the maximum size of a cell is like a two-pizza team and that cell that has a hat and the role of the cell is super clear? So we have, I don't know, at Weaviate, I don't know. Mi is the head of customer success. So it's very clear what she and her team, what they do. That's like they're doing customer success. And the reason it's called a cell is because it grows organically and at some point it splits and they have two cells right and then with a new hat and those kind of things.

Bob van Luijt: 17:12

Now, how is that related to your question? I had a hunch like a feeling that this is the best, that this was the best way for the culture we were building and the way that the company was growing, to structure the company. And when we did our series A, I had capital to actually do that, to execute on that, and when that worked, I was like, ah, I understand that this makes a ton of sense, and so that was my how long, Because I saw it in action. When I saw it in action, it was like this works.

Dr Matthew Mirman: 17:47

What is something that you think that is particularly shocking about Weaviate that you think if the world knew this and behaved like this, you know developers would be working faster start off, so it would be making more money.

Bob van Luijt: 18:04

So the answer sits somewhere, so I'm thinking out loud here. So about what I like to call AI native applications. What I mean with that is that we see that a lot of existing applications where people apply the models and the database is that they kind of sprinkle it over their existing applications or there's like a status quo bias of like this is how we build applications or this is how we build apps. The combination of using these models with the data, with the vector databases, is it's a completely, it's a new paradigm. So I sometimes try to invite people to just lean into that paradigm shift and it's sometimes very hard to articulate for a specific use case or a specific case or what that means. But I'm trying to invite people like just double down on the opportunities that you see here. And this, for example, has to do with we have a concept in Weaviate that we call generative feedback loops. So basically what you do with Rack or retrieval generation, is basically you have a query, you select a bunch of candidates, you shoot it into a generative model it's a little bit primitive now because it goes to the prompt, but it kind of works and then you give the model prompt to do something with it. So there can be. I don't know if you have reviews, they say, ok, show me all positive reviews. You get them out and then summarize them. For me it's very similar a simple pipeline.

Bob van Luijt: 19:45

What a generative feedback loop does is that it stores whatever comes out back into the database with a vector embedding. So an example that we have is that we have Airbnb data and that is has incorrect or missing descriptions. So we have the price, we have the location, et cetera, et cetera, the name of the host, whatever you. So you query the database, you run it to the model, generate a description based on the information for the listing, you store it back with a vector embedding in database. And now if you search for I'm looking for a great place in New York to walk my dog, then it shows something near an essential part.

Bob van Luijt: 20:23

Now, this is often the power of this sits actually in giving a prompt to the generative model to just do something with your database. So, rather than having a human agent interacting with the data in your database, you tell the model you have access to the database. I give you full crowd support on the database. This is the problem that I have. Good luck, please solve it for me that paradigm shift. People are so still hung up on the traditional ways of organizing and structuring data. They say, no, you can't, just, that's just status quo bias, you can just let go all of that, let the model solve it for you. And I hope that that meant to answer your question, that that is something that in this year I had the year 24 just started that that big aha moment for developers or people doing data management or analysts or whatever you. That aha moment comes the model really becomes an independent agent and I mean the generative model to interact with our data.

Dr Matthew Mirman: 21:37

I have two comments on this. There was a hackathon submission winner that won in a generative AI event in San Francisco and I think what they built was a back end server that was literally just an LLM no database, no way of figuring out the PHP. In came the request the LLM processed the request like kept the giant context window or request that had ever been sent and then like hand constructed, like the react. I thought that was pretty cool and the other is more of a question. So right now we're seeing LLMs become actually huge and the context window sizes that you're able to use with LLMs is also really it's growing quite fast. Do you think that that's a viable alternate method to do essentially what you're describing? Just grow your context window, just scale indefinitely?

Bob van Luijt: 22:45

So before I answer the question, may I quickly respond to the first point that you made.

Bob van Luijt: 22:50

So what I think is so important right now is that sometimes they ask me to be a judge at these events and then I'll go and it's great and I'll go there and I recently also did one in San Francisco. So then, as a judge, you just see all these, what people are building. And to my previous point about status quo bias, 99% of people are building something that already exists, but then they try to do better with an LLM. The example that you just gave is a wonderful example of really trying to do something new and I always tell urge these people with these and the majority of people at these hackathons they also try to invent a business, like in the weekend, right over the weekend. So we got to use the element to just run a new startup or create a new startup, and I always urge people just use that weekend to do something super weird or creative, something artistic, something, something far out. So the example that you were just I remember that I saw a video on YouTube like a year ago or something, where somebody was just sending SQL queries into the prompt and it's just like storing data, doing a select star, and then actually the model gave back the right answers. That kind of created. That's fantastic. I love that kind of stuff. So that brings me to your second question. So is the large context window the next answer of storing data in the model?

Bob van Luijt: 24:30

Maybe, and what I do think is important and this is a little bit on a meta level is that at someone, research showed like, well, actually, currently, the large context, when it's don't work very well because there's like this dip in the middle of the data that we put into the context window, that it's like it starts to hallucinate more with whatever. So it's like I forgot the name of the paper Exactly, well, exactly. And a knee jerk reaction that startup founders or people that start off and have like, oh no, look at this paper, it doesn't work. So just, you know, I really think that's the wrong mindset you constantly need to keep. You know, keep looking forward, and if these kinds of ideas emerge of like, can we do that in context window and such a problem here, just somebody will solve it one day, right, and then as a company, you just need to be ready to anticipate on that.

Bob van Luijt: 25:37

So I remember that I was very excited about the research that was happening with I believe it was called MEMET, where they might be wrong, but what they try to do is basically that they try to in real time, they try to change the weights. So, for example, if you had, like Michael Jordan is a basketball player and then they change it to Michael Jordan is a soccer player, and then they ask the model, what you know, what sport does Michael Jordan play? And then it says, like soccer turned out that that doesn't work very well because you're changing, you're basically putting a lie into the model, if you will, from the world, the world representation it has on the moment that it was trained. Because if you now ask the question, who won somewhere in the 90s the NBA cup with Scottie Pippen and Dennis Rotman, then the answer is Michael Jordan. But if Michael Jordan is a soccer player, that can be true.

Bob van Luijt: 26:48

So the point I'm trying to make with this is like a lot of that kind of research is happening of like how will we manipulate the information that's stored in the mall as close to real time as possible? Is it going to go through the content window? Is it going to go through manipulating the weights? Is it going to go through some form of embedding representation? What have you? And I think we need to look at all these kind of things and I somehow was hoping that there was something to the whole memething and turn out that it's not. But we need to keep looking like what's happening right now and as a startup, you need to adopt and do whatever is happening in the space. So I don't know which one will win, but time will tell and it's just super exciting to see what people are doing to try to solve the problem.

Dr Matthew Mirman: 27:48

I think that leads to a very interesting question, which is how are you preparing for if this other technique solves a lot of these problems that we've been able to solve?

Bob van Luijt: 28:01

And so, weirdly enough, I'm not too concerned about that because it's not even weird, it's actually. I think I can make quite a logical argument for it, especially when it comes to an academic way. It's not optimized for these production use cases. So something new happens. Something new thing emerges how you can solve this problem, and it's rarely the case that it's like completely unique, that tomorrow, that somebody would publish a paper tomorrow that is so novel, that is so unheard of. That almost never happens.

Bob van Luijt: 28:57

It's always this build on top of existing building blocks. So I would not be surprised that if the answer is not purely in vector embeddings and how we store and represent them now, if that's the case, then it's not going to be like miles away from how we've had currently is working. So then we might need to create a new type of index, we might need to create a different type of representing the weights or the embeddings or what have you. But we just need to keep an outlook on what's happening. But I'm not too concerned that if we observe it early on, that we can adjust our course, if it happens at all, because we don't know.

Dr Matthew Mirman: 29:41

One thing that I am particularly excited for, or I think that we're going to be seeing in the future, is having these vector databases in the loop and fine tuning or creating.

Dr Matthew Mirman: 29:57

I'm actually a little bit shocked that we haven't seen it. We've seen instruction tuning, where you have people like an army of PhD students write a text about what to do and then solve a problem, and then you give that to a model. We also. The techniques that ChatGPT are based on is literally people going through the text and inputting links for how it should be, using an API call in the middle of a text. I'm just wondering why don't we do that with the ragged in the loop? What is something that you're particularly optimistic about? Some technology that we're not currently seeing that you think that people are going to start working on.

Bob van Luijt: 30:43

I 100% agree with what you were saying. That's also my answer. This is what we call these genitive feedback loops which you just described just having the vector database as part of the basically automating the training process, if you will, or augmenting the training process. We would call it a form of a genitive feedback loop.

Dr Matthew Mirman: 31:09

Of course you needed to take the derivative of your database.

Bob van Luijt: 31:13

Sure, I didn't say it would come in challenges right? One of the things that I've learned is that, if you can so take something as take the example of really in the conversation just doing a vector-based search, just that. Just say, okay, I have a tiny data object and this tiny object contains two words I, full and tower. I'm going to show you on true vector space, I'm able that, if I search for Paris that has a closer relation to the Korean and London For me and people in the space back then, the small groups, people traveling around the world, this took us years to show people how that worked.

Bob van Luijt: 32:04

So, with these kind of new concepts, right, where we said, okay, we're going to have the model and the database working together in harmony or, as we like to say, weaving together, that's going to take years for people to really for the majority of people to understand how that works, because people like yourself have a very unfair advantage in understanding how these things work. But you do not represent the average people creating these kind of solutions with vector database and models. So we need to help people and we need to show people how you can do that. That's why we have a relatively big developer growth and developer relationship, to show the world. These are all the things you can do with it. And then you keep repeating, repeating, repeating, until there's this, until the hive mind has its moment of epiphany, right, and then all of a sudden, oh, I get it, but we're not there yet I read that you need to repeat things when you're just like I'm learning how to be a manager for my first time.

Dr Matthew Mirman: 33:18

It's super weird, coming from academia, but I read what is it when you're like telling people on your team things. If you're just like I think that the website would look good this way, I want people to send more positive emails. You need to repeat everything eight times before people will hear it.

Bob van Luijt: 33:38

Oh, I think if you only have to repeat things eight times, then you've done an amazing job in hiring, because it's often a little bit more. And that's not because, I mean, we all work with smart people, of course, but it has to do with the and that's something I'm actually fascinated by is that these ideas that we have in our head and then, if we translate into language that we lose so much context it's so hard for, and that's also why I think it's important to have retention of people working at the company, because it's only over time that people start to see that context. So, saying it eight times is great. I think it's more. It's close to 800 times, right, and that is because people need to understand all the nuances of the context.

Bob van Luijt: 34:28

So why do we use that type of language in the email? Why do we want to have those types of colors on the website? That is extremely complex. Why do we design the API the way we do? Why do we? And so on and so forth. So it's about consistency in repeating it over and over again and just simple, clear language.

Dr Matthew Mirman: 34:56

I mean, but yeah, if it's only eight times for your team, you gotta think about how many times you have to say it for the rest of the world?

Bob van Luijt: 35:02

Yes, no, exactly. And so one fascination that I have is that you have this diffusion of innovations model and I don't know if you're familiar with it, but it's like you have the innovators, the early adopters, early majority, late majority, and so on and so forth. I am so it's like an 80-20 distribution right and it's bell curve and the bell in the middle that I'm interested in. These people, those are just developers. People just have a it might have like a nine to five job as a developer and they work at a bank or wherever they work. Right, they just want to build cool stuff with it.

Bob van Luijt: 35:47

We need to help these people. It's our task to help these people be successful in what they're building, and one way of doing that is clear content and repeating it over and over. Examples showing this is how you do it. Let us help you be successful. That is. That cannot be underestimated, how important that is. And in our bubble we end bubble being the people who are like working in this specific case, in ML, slash, ai from the early days. We know so much, we have so much context that we sometimes forget that there is this huge group of people who just don't know and we need to help them and we also at risk of alienating these people, right? Because by using complex language and those kind of things, simple language, repeat it, show it over and over and over again.

Dr Matthew Mirman: 36:47

That play. So what's another non sequitur, what is something that is particularly unique about the VVFactor database? I know that there are other vector databases on the market, so what is something that you think really sets you guys apart?

Bob van Luijt: 37:03

So, that's actually related to the answer to the previous question, and this says like so let me give you one example. So we have the modular ecosystem, because the question is, these embeddings not for everybody. It's easy to create them. So it was like oh, you know, if you use a third party service like OBI or career or whatever, you just fill in your API key database takes care of the rest. Or if one of the people want to run their own models, those kind of things, just to make it easier for people to work with the models in combination with the database. Another example that comes out of that is that the majority of the vector databases take as input only vector embeddings. You can do that with VVF2. However, for if you want to do hybrid search, you can also send in just text queries, and one of the things that we start to notice is that for a lot of developers, that's just easier. Oh, it's actually easier for us to just throw text at it and, yeah, apparently it creates vector embedding in the background, but you know, I'm not aware of that. And so, to answer your question, I think lowering the barrier for people to start building end-to-end applications with the database, that is something that sets us apart, just make it easier to build end-to-end applications.

Bob van Luijt: 38:30

So I think there is a case to be made early on for being a little bit scrappy. So if you make a demo, for example, so think about the situation that you are watching. You're sitting in an audience and somebody is giving a demo on stage of a completely new thing. Just take the example you just mentioned of somebody that we do a sequel inside the LLM and it's such an amazing thing and you show it, you set it correct and the thing goes wildly off the rails. You get some laughs, but it's also real and it's new. So there is a case we made to be a little bit scrappy now and then. So it's a because the danger there's a form versus content danger. And what I mean with that is if we create infrastructure and we, being like anybody who creates infrastructure or any deep-tech product for that matter, we have the content, so what we're actually building, and we have form around it. So it's like how we talk about it, how we show it, how we market it and those kind of things. And I see startups now, especially what's happening in AI, who are very, very good at the form side of things. But then if you look what's the actual content, then you're like ah, one could be critical about that.

Bob van Luijt: 40:20

The problem is, in my eyes, is that if there's not a healthy balance, it can also be the other way around that you build an amazing product, but you just don't know how to show the world what you do. And the problem that you're dealing with is that it's an unhealthy balance that hurts the culture, the people working on the product and those kind of things. Because if it's balanced the wrong way around, that it's just only like product focus, content focus, and people might go like, yeah, but nobody's buying it, nobody wants to have it. But if you do it the other way around, they're like we're selling crap and if you're early stage, you just don't have a lot of money. So you need to be a little bit scrappy and there's something to it and I love that.

Bob van Luijt: 41:12

Also with a DEFRAEL team and they have drones and stuff and they fly around. But it's not like a Hollywood level drone flying Developer flying with a drone filming that. I'm putting that on LinkedIn and I think that's very cool. That is super cool because it gets people into like, oh, this is the start of this, what they're building, we're excited for them and that kind of stuff. So I don't know what you show, but it should be too bad, but it's a. There's something that if it's too polished, if I get a deck from a startup that's too polished, I'm like so when are you building products?

Dr Matthew Mirman: 41:55

So that's balance. I'm an anecdote, for that is at the very beginning. I mean, this is not my first startup. So in my first startup, the first version of this product that we built looked a-janky and all of my friends loved it because it looked a-janky. And then we spent six months, we made a much higher quality version of this product. We put a ton of engineering in it, but we were just out there and talking to people at the same time and then we did finally release. We were shocked because everybody was just like no, you're a big company, you were evil. Right now. I was like what? Three people in a garage. So the second time I built the company, the first thing that I did was like our demo product that I insisted it's got to look like a 90s, like the Ladox website, because who could have hate a 90s Ladox website? Exactly.

Bob van Luijt: 42:53

But I agree with you. It's like a if you build infrastructure, the first demo should be API calls, Because what you're signaling with that is that, first of all, the audience you want to reach are people who understand what you're doing, and if you do something new, that's like a small group right, those are the innovators. By showing it from the perspective of the API, you show there's no fluff, it's just API calls and this is what the thing does that comes out of them. That later to reach the majority and that you need to help them with graphical user interfaces and maybe, if you want to sell into the enterprise, that you need to have certain content on your web. Sure Not when you're early stage. When you're early stage, you indeed want to have those kind of apps. I saw the recently. I saw the website from a company called. I believe it was like well, I mean, your website actually is a good example, but because I'm blank on the other name but your website is a great example.

Dr Matthew Mirman: 44:01

It's like you see this is a startup building something new. Did you just call our website jinky?

Bob van Luijt: 44:10

It's sort of it's in a good way, that's good, that's what you want. I remember when we started I thought that purple was beautiful color, so the whole website was purple and the first Defrel I hired Sebastian, his first. We had the first conversation, he's like so I looked at your website. It hurts my eyes, man. There was the first thing he said and he was you know. I said he was right. But that is that it's. It's that's part of and that's for him was interesting. It's like okay, there's a lot of work to do here, so it's a in this, in a that's in a good way, it's a. There's just something weird about a that is like a like that's like a child in a suit.

Bob van Luijt: 44:58

That's like I mean you know, it's, it's, it's funny to have a child. It's kind of weird, right, because he just wants to run outside and play and do whatever.

Dr Matthew Mirman: 45:07

But also you know that, like six year old, that you knew that wore a suit to school every day. Like he's probably off, he's the CTO of IVF or something.

Bob van Luijt: 45:16

Exactly, exactly. It's a, it's a, it's just there's something awkward to it. So it's a. I don't know where I got the with the metaphor, but the there's something. There's something, there's something off. So so there, a tech startup needs to be a little bit scrappy, just a little just sprinkled over it.

Dr Matthew Mirman: 45:34

You also have to figure out how to take that like extremely small amount that you have and turn that into something like crazier and turn that into bigger and find all of the acts to multiply whatever you have. For example, like this setup, like normally you want like a nice podcast setup, you have to go and hire an entire team, travel with you. Each camera has a person standing behind it. Like you have an entire production crew and like, but we're a four person startup, so like yes, yeah, so I, that's something I call the wine problem.

Bob van Luijt: 46:08

So it's the. So when I went back in the days when I was a student and and if he, you know his friends bottle bottle of wine, then we bought wines that were like six bucks, five, six bucks, and then during the weekends and we just you know, maybe we're like it was like weekends like okay, let's go for an expensive bottle. So now we didn't buy the six bottle wine with like the 12th, you know a 12 bucks of wine, and it was that that one was definitely twice as good as the six. You know less of a head, that kind of stuff, right? So, but I mean you can buy bottles of wines like five, okay, you can spend five, 10 K on a bottle of wine, and surely those wines are better than the, the, the, the 12 dollar wines.

Bob van Luijt: 47:00

However, that it that doesn't grow in a linear fashion. So a a six dollar sorry, 12 dollar wine might be twice as good as a six dollar one, but a 24 dollar wine might not be twice as good as a 12 dollar one. So, but the problem is, although the distribution of linear right, it's like a at some point is, it gets harder and harder to actually. You know, it's flat and it's right, so it's harder and harder to see what the but if you want to do stuff better, um, um, um, uh, you need to. You know, at some point, if you, if, if, if you, your podcast becomes like a wild six, at some point you will hire somebody to help you with these kinds of things, right, but so then it becomes better, but it doesn't become linear better in linear fashion, it's not. I mean, you have a beautiful setup here, right, but if you, if this now would be humans with handheld cams doing it, I'm not sure that the podcast would be twice as good. It would definitely be a twice expensive at least.

Dr Matthew Mirman: 48:03

Yes there.

Bob van Luijt: 48:03

Yes, exactly, there's like just diminishing returns, right. So NZ and that's like a um, and I always I call the wine problem because it's like a visual way of to visualize it. So what?

Dr Matthew Mirman: 48:13

the same things have you seen people do with VBA oh?

Bob van Luijt: 48:17

yeah, I have, so I can. So I, um, how should I tell this story? The um I was once in was in the amount of you, and I met these three guys at an, at an uh event. It was actually organized by Google and like, hey, you're the review guy, we're a review user, yeah, great. So, um, and we were there for the models that um, that were presented by Google, and they said, no, we store everything in VV8.

Bob van Luijt: 48:54

And then they had this whole thing that they were not using Google's models. So, um, so that guy from Google asks so why are you not using our models? Why are the open source models best? No, then it was a very fake. He was like no, no, no, no. So you know, the thing we make is we, we can do that through like APIs. We can only do that Like if we host the models ourselves. You know why? So they were like so what? So what does your product do? So turned out that they built this NSFW bot. Then they were storing the data in VV8 and they had these open source models because if the actual text, if they would send it over the API, they could just got back like a like, like an error message. They just couldn't do that Right, and we were sitting there and I was like, oh, okay, thanks for sharing.

Bob van Luijt: 49:43

And that was like the most, the most, I mean the most awkward thing that I saw. Somebody. Somebody built like we're like, oh, but it was super funny, but it's the the the nice thing is that the um, that there are always these little funny things that people are building. Or I remember, on a more serious note, that very soon on some um, that NASA published this paper, that they bought something with VV8 to analyze like data come from satellites and stuff, and it was kind of it was really cool because that felt so, so, so, like a grown up use case.

Bob van Luijt: 50:22

Right, it was like it was like years ago and I was like, oh, really, did you guys build a review? Yeah, we did. And I was like so that's, I mean, it's like maybe like a weird orbit, just get. But it was more like a um. That was really different. So and I enjoyed that a lot. But the um, when you asked that question, that story came to mind and I was like, and we were, everyone was sitting there I was like, oh yeah, gotcha. So that was very, very funny.

Dr Matthew Mirman: 50:44

So when they were hosting your NSFW conversation were they? Were they hosting it on your like hosted platform or they hosted?

Bob van Luijt: 50:52

themselves. They hosted themselves yeah.

Dr Matthew Mirman: 50:55

I guess that leads into the question of like do you have content moderation policies for your hosted platform, and is that even something that is feasible if you were building a public database, something like you know super base or copper robe shirt?

Bob van Luijt: 51:15

So that's a great question and um um, believe it or not, I get similar questions sometimes because you know, with the models and those kinds of things, and sometimes they're more like in the ethical realm why, and um, um? So currently the answer is no, we don't. And so with open source, you don't. You don't know what people do, right, so that's like impossible to to, to check. But you manage platform. Theoretically speaking, you could, but you don't want to do that because you're not allowed to do what people are actually storing inside the, uh, the database.

Bob van Luijt: 51:53

So the way that I currently deal with these kinds of situations is is um, um, you just go on your use of a moral compass, right, and so if, if you see a company, uh and I, we, we didn't have any issues with that. But if you see a company sign up, where you go like how was this or what are they trying to do, you try to figure that out and then you just go on your moral compass, but the um, um, that that goes for any ethical question. The people can have around the data that they're storing in the database or that they're generating with the models or those kinds of things. I'm very aware of it Um, uh, but luckily the, the, the, the 99.99% of the use cases are very um uh, you know, uh are just normal, regular use cases, but the we don't have. I, I, I would feel very bad to put a filter in my uh, behind my API to you know, to to uh try to figure out what people are storing inside the database.

Dr Matthew Mirman: 52:54

That's none of my business when I say you're ever gotten into a situation where the Taliban has come up to you and asked like, can we use, we be it.

Bob van Luijt: 53:03

No, I've, I've not, I've not been in that situation.

Bob van Luijt: 53:07

What they, what they, um. But I mean I, you, you do get in situations where you know there's like certain countries where, um, because we've been, we just literally use all, almost all over the world. So, if you, if we look, if I look at my website statistics, right, so almost the whole map is like blue and that's the color of the, the tool we use, uh, with with with a few obvious exceptions, like I don't know North Korea or something, right, so so there are countries in there that have, just from a governmental perspective, so they're like sanctions or those kinds of things, and and I have been in situations where these companies reach out and it's just in a very normal professional way Okay, we, we want to, you know, we use your technology, we're going to buy one of the service, but that's, that's tricky because, um, we're just a startup, right, I don't want to get into issues just selling to, to, so it's a that that did happen, yeah. And then we, we said sorry, but you know can help you, but you can use the open source technology, so that that has happened, yeah.

Dr Matthew Mirman: 54:08

Do you think you would be confronted with all of these ethical challenges as frequently if you weren't, in general, a AI?

Bob van Luijt: 54:17

But no, absolutely not, because I cannot imagine. It's hard for me to imagine that in the early days of MongoDB that they got these kinds of questions. So what are people starting it? I that's hard. Maybe. Maybe they people did, but I it's hard to imagine. But you know what the thing is. It's also and that the kind that also annoys me a little bit. There's like these naysayers and you try to do something new and then people ask this kind of question. I once spoke at a conference and there was when we it was like 40 people or something and I got a question from the audience. If I had a something like a question do you have an ethical board? I'm like I'm a 40 person startup. No, and and um, I was completely surprised about the question and the answer to that is yes, I'm an ethical board.

Bob van Luijt: 55:07

Exactly it's like we have. Yes, exactly, but I was like we have a moral compass, that we, that we just, but no, of course I don't know. I mean, we're building a shipping product, that's the focus, and um, uh. But that person was an academic, should send me a long email. It was not happy with my answer, but I, I, I, I said like, I'm sorry, it's like it's, it's the, we have a moral compass with the, with the company and that's, and, and, and we try to do the best we can, but we're also just shipping products, please, right, and uh, but for some people it's a, it's a day, some people don't, um, some people don't see that. So that is, that is, that is like.

Bob van Luijt: 55:52

Sometimes you see that with young people, when they go and work for a bank or something, right, and then every month they get their salary and they kind of assume that that money is always there. And especially when you want to start up, that's not the case. You need to build and ship great products because otherwise it's just not going to work. So that is priority number one, two, three and four at the same time. Right, and so if you get in such a situation, you deal with it when you get in the situation and if you grow bigger as a company and it's a recurring issue, again I mean doing wrong if not. But I'm not going to be. I'm not going to hire people to think about that If it's not a problem, especially not in now in the 16 people company, but I do get those questions way more often than if it's not generally related. That's a good point.

Dr Matthew Mirman: 56:49

So what was getting? Your first investment like Um.

Bob van Luijt: 56:53

I. The first investment that I got was um. I did live um in Holland. I did. We were part of small accelerator, but I'm if I'm coming in actual investments. So there's our seat um round. That was from Zera in services go I re.

Bob van Luijt: 57:12

I was on a holiday and it was in between COVID. So it was like uh, when there was like um, it was like a half lockdown where you could go to hotels and stuff, but there was also kind of a lockdown happening and I was with my brother and my dad where he was on a short holiday in France, as I think the investment is going to happen. So I was at my phone open and and I was the, and I was um, uh, scrolling down and and and we had like almost no money and um, we raised uh, 1.1. And all of a sudden I was like 1.1 million dollars on the bank card and I was like I never seen so much money in my life and um, and, and you, of course, um, um, um, asking what it up to that. But the reason that I started with this was because I was so proud in that moment. I was like we had, we have this idea, we're working on it and somebody is willing to give us money to explore this further and work, work on it. I was so proud on that moment and what led up to it was the um, uh, this was the only time in my funding history that I had real bitch decks. I saw that a bitch deck and I had to, you know, and it was like you know, shopping it around and showing it to people.

Bob van Luijt: 58:33

Um, and after that, with everything happened in ML it. That went always smoother, but that was the first. I and we started the company. It was like we're going to try the year and if it doesn't work out, it's going to quit. And, uh, so we started. We didn't give ourselves any salary or anything. So we started working on we've yet, and, um, pretty soon actually, the first conversations happened that I think we were like maybe half here in or something. And then, and then it became very serious so the clone, this is nice, this, this works. So, um, um, it was a, I think, and to answer your question, the best way to answer is I was so proud, like somebody is willing to give us the opportunity to try it out and see if it works, if we can make it work, and so I was super proud. That's the. That's what we were answering, I guess.

Dr Matthew Mirman: 59:30

I mean it means so many folks. Yes, I feel like there's a lot of like in the bootstrapper and in the like, even the YC circles. There's a lot of anti-fundraising going around right now and people are like, well, you shouldn't get validation of it, but it is validation. You know, like for a lot of people, they never built startups, like even in their own communities. Like, especially if you're the sort of person that's going to build a startup, you have a weird ideas and everybody around you like will often think your ideas are weird. So, finally, somebody's telling you like your idea is not that weird, like this could actually be huge.

Bob van Luijt: 1:00:09

Yeah, and there's something else also very important to mention, and that is the. So that's just a bucket of. If you want to take VC money, because it's indeed as you're saying, it's actually you don't have to right, it's like a, it's optional and you don't have to, but if you believe it's needed, there's like a bucket of money available. And when you work on software, it's an infrastructure that mostly gravitates around the Bay Area, and I mean it's another place in the world, of course, as well. But I mean it's like you know, if you want to play in the NBA, then you know, then you have to go to the Bay Area, and so these, you see, they somehow need to figure out, like, who's telling that we want to get money to this? You can't show numbers yet, as in, like, you know, revenue numbers or the. It's all very early stage, right, and so the, the, the bond that they fish in is students from Stanford, berkeley, people who worked at, you know, the, the, the fan companies, those kind of things.

Bob van Luijt: 1:01:24

I had none of that. I was living in Amsterdam, I've studied music and I was writing software for a living. So fighting your way through that and then, in that way, getting that validation was very important to me and the because I didn't. I didn't have that. I couldn't show I didn't have a, a Stanford you know diploma or anything. Or I could say look, I worked for four years at Google, I didn't have that. So then the only thing you have to show for is your is your project and what you're working on, and that you are a fighter and that you're going to try to make it work. So that, and that's a. I mean I do think it's important, because some people give up or any in that stage before because of it. How can I ever become part of it? And the answer is I'm an example. You can just keep going.

Dr Matthew Mirman: 1:02:20

How much did you guys have in the bill at that?

Bob van Luijt: 1:02:21

point. So the the first first, we had a dependency early on on Elasticsearch and so we because we decided to build top down so we started with the APIs and then Elastic, of course, based on the CNET no vector index so we had the vector index on the side and then my co-founder, hssec if you really want to do this well, we just need to build an actual database If you really want to do it well. And when we got the seed funding, we still had that dependency and but before we exhausted that money, that it was 100% standalone.

Dr Matthew Mirman: 1:03:06

So I guess when it started it was an open source.

Bob van Luijt: 1:03:10

Well, it was actually so. It's like. You know, it's a funny story. So if you I started myself and I was experimenting with these vector names, I would also experiment with different languages. So if you take the Weave-Ed-Repo and you really go back to the first commits, you see it was written in different language and all of history is still there. So Language was it in Node Nodejs, because I hadn't foreseen that it would become an actual full end-to-end database back then. Again, we're talking now. It's like 2017 or 16, 17,. It's a long time ago, and but I did know something with these embeddings is funny.

Bob van Luijt: 1:03:53

There's actually a Google podcast. It's called StackChat. It's on the, on the YouTube, the Google YouTube channel, where you find a interview with me from way back, where I talk about Weave-Ed, but in its previous form. So we talk about the embeddings, we talk about search with embeddings, but how I talked about it was different, right, and how Weave-Ed-Benguin was built was different. So that's the. That was the early version. But before we did our series A, we had a. Why do you switch off of Node? Oh, it is very easy, because the, the, when we figured out it needs to become a database, we needed to have a language, that in which you could actually build a database.

Dr Matthew Mirman: 1:04:39

You know, I've never seen a JavaScript database. So now we're thinking about if JavaScript's one of the most optimized languages. So you don't exactly have access to pointers, but I wonder if you could make it work. No, but that oh.

Bob van Luijt: 1:04:50

I mean I mean we have, we have now database competitors that written the whole thing in Python, right, but the problem is at some point you need to go to a, you need to go to a lower level right Of the stack to really optimize for larger use cases and and you're just gonna get stuck with these kind of languages. At some point you're gonna get stuck and that's why in the most database today are still written like C, c++, rust is now a thing, go. That's why you see those, the older ones, still in Java, because it's a. Otherwise you just can't do it. And I mean this is beyond, out of my realm of expertise, but but I see I mean.

Bob van Luijt: 1:05:42

So, for example, we have certain optimizations in Weaviate that are written in assembly and that's just. That's just easier to deal with those kind of languages, right. So it's a. You could theoretically do it. I mean it's interesting. It's also the question of what is a database, right? So that also plays a role in, but at some point, like from a market perspective, we making money from database perspective often sits in the operations. So you wanna have the right controls to operate, to create operational excellence in how you wanna operate the database, and there's very hard to do with languages like Python, javascript, those kind of languages.

Dr Matthew Mirman: 1:06:27

That is actually a great question. What is the database to?

Bob van Luijt: 1:06:30

you, yeah. So it's funny that you asked that, because I was when I said it was like oh boy. Now you're, of course, gonna ask me what the definition is. So for me, the definition I look at it more from a from a, from an DX, from an experience perspective For me it is cross support, so you can create data objects, you can read it, you can update it and delete it Efficiently and reliable.

Bob van Luijt: 1:07:01

So that means that if I create a data object or row in a table, whatever, and I leave the database B for a week and I come back that it's still there, that I still can retrieve it or that I can update it or that I can delete it, and that is, for me, a database. That's it. It's like reliable and persistent, that's it. And then we have all these flavors on how we can do that For all these different use cases that are around and it can be different indices. So when you do a search, for example, the effect right next there can be separation of storage and compute If you have more OLAP kind of use cases. But in its core, that is what I believe a database is.

Dr Matthew Mirman: 1:07:48

How have you seen AI change since starting with UVA?

Bob van Luijt: 1:07:52

Well, first of all, we didn't call it AI, we just call it machine learning. I was extremely hesitant to use AI on our website. I was like, really, really, our people got to buy that as in like, but we had to. So that's the change that I've seen, just the use of the term AI. That was okay to say that. What I've also seen is the how should I say this? So if it's like, if a gold rush is happening, it attracts certain type of people right To go, like who's trying to figure out how can I, you know, how can I make some money out of this? And in some way, I'm guilty of that wave as well, in the sense that in the early days it was, of course, only an academic exercise. So I was already one wave after that, when we were more like, hey, we as developers or entrepreneurs, like, hey, we can do something with it. But then it started to change to people who were just trying to just make a quick buck and that kind of stuff. The, everything that happened in cryptocurrency, was not helping. So when the, the, the, the vehicle, with, with FTX happened, all of a sudden, the you know, the crypto boys became, became AI, you know, then became the algorithm. All of a sudden they also knew something about AI and I was concerned about that because it was like we are working on with our peers in the space of course, that includes our competitors, but also our partners, et cetera. Just to the community at large. We really try to make this into a thing that people can use and benefit from and can build new business on top of.

Bob van Luijt: 1:09:47

And I was really concerned that with the, the, the, the weird, I mean, remember there was this guy who sold AI courses and then he was like any any church, a hundred bucks for it. And then he had something like, yeah, solving the stock market or something with AI, and they had people pay for that. I was like, oh yes. And I was like, oh boy, that's not good, because because if people see that, if people start to see too many of these scams before, as an industry we really have shown the added value we can bring to companies and we're now talking, yeah, like one and a half years back or something. I was really concerned about that. So that's a big change that I've seen, but it's kind of it. Just it comes with growth, so there's an opportunity and people jump on it so. So I was concerned about it. I wasn't, you know, I wasn't. I was fine with it most of the time and it was like I was concerned that before we actually got to where we wanted to be, showing how we're really bringing value in this new way of doing things, but that, you know, knock on wood, but that luckily didn't happen, but it was like that was concerned about that. So that is a big change that I've seen. Also, a change that I've seen and that is something that's really awkward to go through yourself so is that Innovators dilemma was absolutely happening with incumbent players and, for a long time, the question of where are they?

Bob van Luijt: 1:11:25

We have these new startups, we're doing this vector search stuff. Where are they? And what happened? Because AI was growing.

Bob van Luijt: 1:11:32

Yet all these companies I mean Google, is a good example. So two years ago, no, I might be, or a year or two years ago, I might be, I might be wrong, but so Google, I always happening and they say, like you know, you can use our at the keynote. They're like not, not a breakout, such the actual keynote. They go hey, you can use these embeddings. These are the partners we work with and we, as we've hit were on the screen no big existing database players, because they didn't offer that service yet.

Bob van Luijt: 1:12:07

So there was this time for over a year that these big company Intel, intel does a keynote and they show what you can do on the Intel cloud or something during the actual keynote, or the CEO is to and the presenter, like I, built the following with Weave Yates and he goes I mean, like what really, and all these kind of things were happening with all these big players, that that because the, the incumbents were just not there. So so we, we just could keep building the build customer base, build community, build product, et cetera. Well, there was just, there was just it was just a green field and that was something that was really fun. But also, yeah, kind of where to just observe that I was, I mean, don't I was so proud on all these and all these things, but I was like now, if you go to the, to them from the Gemini website, from from Google, right, so it's like was it AI, google, that ever something? This is big, we did love. I'm super proud of that, but I could not have imagined that that would ever happen.

Bob van Luijt: 1:13:14

So that has changed that. He that that, that there was that because people started to see what the capabilities were of AI, plus the fact that there were only new players in the space. That we just could. We just, you know, we just stepped right into the opportunity, we just didn't have to do anything, we just walked right in and and and that shows what a systemic shift is happening in space. Because I have a friend who is a friend of mine. I have a friend who has a leadership position, had a leadership position at Google, and he's like do you know how? How abnormal it is that Google reaches out to you to ask if they can use your logo, rather than you asking them where they want to use the logo. They said that is that. That's. That hasn't happened for years. And now this whole AI thing opens up and then goes to one example. There are many of these examples, so that is something has changed and I'm super proud of that.

Dr Matthew Mirman: 1:14:15

So, so, so that I think that's something we definitely something to aspire to, yeah.

Bob van Luijt: 1:14:21

Yeah, that's also to do with so, but this is also the. What's fascinating in building startup is that the there is this and again, for me it's I'm very, very, very aware that that we also got lucky right that and with lucky I mean that who knew? Who knew that it would take off the way it did, right? So but if you have a startup idea, there's an interesting correlation between If you decide to, you take VC money with the amount of money you raise versus how conservative you become in building products.

Bob van Luijt: 1:15:05

Because if you have no money or you have, I mean even VC money, right, if you don't have any money and you're with your co-founder, you can do whatever you want. You're going to go as crazy as you want. And now I want to apply for VC. Let's tone down a little bit the craziness, because otherwise you're just going to get anything. And then you do VC and you go. Let's tone down a little bit, right, and that is unfortunate. So this is like a with category creation. Then you could argue like just keep looking, keep searching until you found that niche that you can create a new category around, because otherwise you just get swallowed up by the existing ecosystem.

Dr Matthew Mirman: 1:15:54

Your job is to tell people why you're doing something. What is your company mission and how did you come upon that company mission?

Bob van Luijt: 1:16:02

Yeah, so we want to enable and help people to build AI native applications. So let's unpack that. So let's go in in reverse order. Effective database is a tool and people use that tool to build applications with. And I believe I honestly believe, as I mentioned earlier in the conversation we can sprinkle machine learning over existing applications or, with the state of AI right now, we can really build completely different applications for end users to use and to buy. That is what I call AI native. That is where the AI sits at the heart, the fact that it is today a core component of that.

Bob van Luijt: 1:16:49

But I also tell the teams like we focus on helping people building these end to end AI native applications and yes, the database is part of it, but that's just part of it. And then the second thing in that is the word help. So I became convinced that our path to success is by being extremely good at helping people to be successful in the use cases they want to build. So that's not only sits in the database, that not only sits in documentation, but that also sits in that we have modules. That also sits in the video content they were creating. That also sits, that we decide that the language we're using. So the majority of language that we're using is aiming at a broad audience and it comes back in everything.

Bob van Luijt: 1:17:47

So, for example, my co-founder, hn. He gave a talk for the Carnegie Mellon for the database group, but even there, how he talks to the students yes, he talks about complex topics how the index works and charting but the way we do it, the language that is used, the visuals that are used are focusing on let us help you explain how we do it rather than show off how we do it. So those two things combined, so helping people be successful in building these end to end AI, native applications, where we feel is like you know, brand and banner. Basically, that's the mission.

Dr Matthew Mirman: 1:18:29

Is there anything that you want to say before heading out? Any final words to the developers of the world?

Bob van Luijt: 1:18:37

Yeah, have fun building. This is the time right. The last thing that happened was mobile phones. Before there was the internet itself this whole new opportunity to build complete new things. The time is now and have fun. Do it now and have fun building. That's what I want to give people. You know, something to think about.

Dr Matthew Mirman: 1:19:06

I think that's something that I'm always trying to be doing. Well, you are. And then last question, before we have to head out, is there anybody that you want to acknowledge that was particularly helpful on your startup journey? That has been helpful recently.

Bob van Luijt: 1:19:25

So I'm not a given, maybe unconventional answer, if I may, and that is everybody. The further I get in my journey, the more I understand, learn and appreciate how wonderful it is to work with smart people. We, that is, people, the developer relations team, everybody. It's my co-founder, it's the advice that we have, it's like investors that we work with. I'm just. I mean, it's such a lucky position that I can just talk to these people, listen to them, they share their wisdom with me about this specific area of interest of theirs, and the only thing that I need to do is like I need to tie it together and I am enjoying more and more that I just, I don't know anything and I'm just and I'm I'm working with with everybody. So the, the, the answer is everybody is like it's a. I'm learning every day from everybody related to the company. It doesn't matter if it's employees, if it's if it's my advisors, friends, community, my parents, anybody, so people, I'm learning every day. So I would like to acknowledge everybody's helping me, you know, trying to make we a success.

Dr Matthew Mirman: 1:21:01

Well, that's wonderful. Thank you so much for coming on, Bob. Thank you so much for having me, man.