Seltz raised $12.5M seed: Antonio Mallia, CEO at Seltz

0:00Antonio Mallia:Hello and welcome back to the Scaling Europe show. I'm Sid Johnson. Today I'm joined by a very interesting guest with some very exciting news, Antonio. Antonio is founder, CEO of Seltz, an extremely exciting company with some great news to share today. Antonio, take it away. Hey, Sab. Thank you so much for having me here. It's a great pleasure, a great pleasure to talk about Seltz. So what's the news? well um yeah we are just uh finally announcing uh our uh you know seed fundraise we raised 12.5 million um we are we are now launching uh you know our our initial product uh and and you know is a web search company um offering web search for ai and nlms ai models uh the new user uh of the web.

0:52And, you know, I think today is a special day for us, for the team, because we are now enabled to, you know, offer a new methodology to access the web to AI models.

1:09Antonio Mallia:And so maybe for those who don't have context, tell us a bit about why this product is necessary. You talked about agents being the new users of web search. In what ways are they using it? And And why is your product necessary in this new world? Yeah, I mean, you know, AI models have a completely different way to reason and consume content that comes from the web. And as of today, the kind of external and traditional search engines that have been built for humans do not offer what AI models need to maximize their ability to extract information and come up with patterns and come up with answers and tasks that are actually completed in that case without hallucination or with a very reduced and more trustworthy outcome.

2:13And so we're building a different solution from separate angles, from the input. AI models can generate very long and complex queries, not even to mention that they can run thousands, hundreds and thousands of queries in parallel. So the query load, the type of queries are completely different, but also what's really important is the output produced that they will consume is actually quite different. The old 10 blue links doesn't work anymore when it's an LLM or an AI model using it. They need the actual information. They need to know why that information has been brought to their attention. and they can leverage some of the retrieval features that we actually use for ranking for retrieval.

3:05They can use it to personalize the output that is then proposed to the user or used in the AI workflow that they operate.

3:20Antonio Mallia:And what you're building is you're essentially rebuilding the entire stack, right? There are some AI companies that have sort of built on top of existing search infrastructure. my understanding is that you're going all the way top to bottom building the entire stack can you talk a bit about what that means in practice and how it's different to existing companies yeah yeah and i i think this is a very important point to make it's like we decided to go through the kind of like harder part where you know uh this involves rebuilding the entire stack from from crawling to you know content ingestion up to you know how we present the results uh in the API, that's practically a very complex problem to solve.

4:05And for this reason, we have raised this round and we have hired the best talents, the experts and veterans in information retrieval around the world. Building just a thin layer on top of some of the existing technology today doesn't allow you to change paradigm. And so the degree of freedom that you have is actually very limited. So you won't be able to maximize and optimize specifically for AI models. But that's why we decided to took the hardest path. This obviously comes with a lot of complexity. But this is exactly the reason why we are going to be successful. why we are going to build a product that is going to be able to empower AI models in a different way.

4:58Antonio Mallia:And you're now equipped with this$12.5 million suit round led by Speedinvest and B Capital, I believe. Can you touch on perhaps what they saw was so exciting about the company that you're building? What gave the investors conviction to back you and the team? Yeah, I think they were really passionate already about the space, first of all, which is what caught their attention. And when I say they were passionate about this phase, I mean, you know, they probably knew how complex it is to pretty much like do three different things, which is how I spent my day, you know, since I started the company, which is like, you know, hiring, thinking about the product and thinking about sales.

5:43Like, you know, these three things are kind of like my obsession these days. and it's also like what probably convinced the investors we have to back us up. And the reason is if you know the space, if you know how search engines work and how they're important these days for AI models, then you know how it is hard to hire the right team to build such a product, how it's important to build it from scratch, but also how hard it is to find the right people that can help you build this product. And then second, on thinking about the product itself. Coming up with this new retrieval methodologies is not simple.

6:25And reshaping the APIs and interfaces that these models can leverage is really something difficult and having a vision is one of the reasons why we, a vision that we shared with our investors was one of the reasons why we decided to work together. Yeah, and finally, convincing, but at the same time showing value that there is needed to be detached from a wrapper around existing technology is something you have to communicate to customers in a different way. So I think they had also that vision when we were discussing about investment. They were like, oh, this is the product that needs to be presented in a different way.

7:21Antonio Mallia:And when I look at your background and experience, I think you've come from quite a research, almost academic background, that seems to be very much in the space that you're now operating in. It seems like that's probably, when you look at the profiles of maybe some competitors or other players in this space, slightly different, right? It seems like you've actually got a much more research-focused background than other people in this space. Do you think that's been a key reason why investors have been excited? Yeah, I would say this, I think, in the way we are going to be shaping like, and, you know, we are already shipping this new product.

8:01I think this plays a role. 70 % of our team is actually coming from academia. It's essentially like an information retrieval lab where we can experiment with new methodologies. Obviously, you know, the team is composed of people that have built web search engines at scale in the past, but they have the vision and they have fact-report of how to actually innovate when the problem is unclear and unstructured. So it's a completely greenfield project, in my opinion, these days to deal with web search for AI models. And we need talents, we need minds that can go and solve one of the hardest problems that I see that's out there today.

8:48So yes, I think having a research background or having like, you know, tens of years of experience in the space. It's quite important. And that's how I, you know, structure the team in a way, you know, to have people coming from a variety of, you know, past experiences.

9:11Antonio Mallia:What are some of the challenges of having such a research-heavy team? Because I think, you know, when I look at startups, I look at the makeup of how startups have changed, five or six years ago, very few startups would have had a lot of research people in their team. Right now we're in this age of much deeper tech, much more technical technology where you need a real, real academic experience and a real base of knowledge to be able to build upon. But I imagine that the profiles of some of the key hires are very different to who they were five or six years ago. They're probably very, as you mentioned, academic focused.

9:44Antonio Mallia:What's it like working with a team of researchers? You know, is it interesting to, was it like pulling researchers into the world of startup and technology in VC? I was really surprised at the beginning as well about like researchers being interested in joining a startup in a way. Like people that have been, you know, attending CKIR or, you know, academic conferences and publishing in these top tier venues and then being excited about joining a startup. I think things are a little bit different nowadays in the sense that, in particular, when it comes to, you know, search and information retrieval, it seems that because of this paradigm change, there is more interest in academics to kind of like, you know, show what they have learned in a way and to, you know, what academia has brought.

10:39Because when I think about like information retrieval, it seems seemed to be a solved problem five years ago before Transformers. And now things have changed dramatically and, you know, they are drastically different. So I still see a lot of excitement. Obviously, having a team with research focused and so on doesn't really mean we're just working on research projects. We still have a lot of practical implementations and there's a lot of problem-solving happening that impacts customers directly. And I think that's also the other part that really excites the team. I remember having a conversation with one of the team members saying, like, Antonia, I don't really care what problem we are going to be solving.

11:37I care about having an impact to our customers. So that, for me, was fulfilling. It made me excited. So, yeah, I think it's really a balance between the two parts.

11:51Antonio Mallia:can you talk a bit about the traction that you've seen i know that you're seeing a lot of uh use think you've signed some customers can you kind of say where you're at in that side of the journey yeah absolutely um yeah so we are in a position where you know our product is essentially brand new which we just you know launched our first uh um you know endpoint which was around uh news For that specific endpoint, we actually created a benchmark. One of the problems with benchmarks is that usually the company that is launching their product is also the one that is benchmarking. So it's never an independent test that we're running.

12:36But because we saw that limitation and because of our past as researchers, we kind of thought about a way to make it fair. And so one solution we had to make it fair, make it have a fair benchmark, was to generate queries, everyday new queries, so people, and not even us, can predict about the news and the events that are going to happen the day after. You never know, right? So you cannot optimize this system. And I was mentioning this because this was actually a key component that our customers or the pilots we are having with the initial proof of concepts. And they start because they see the value of this benchmark.

13:31There are either people that are really using some of the APIs available on the market, and they see the difference when these APIs or these products have not been optimized towards a set of queries. So it's a dynamic benchmark, so you can really optimize towards that set of queries. And these customers actually realize that that's actually important because the data is always changing, the web is constantly evolving. So they see this value and they decide to try us and compare our product with what they have. There is another set of customers that doesn't know about web search and doesn't know about the advantage and the improvements that you can get if you plug pretty much external knowledge into an agentic workflow.

14:24And so for us, it's really a discovery right now to see which are the companies that are probably not experimenting enough with the new technology that we can offer, the ability to inject engineered context that comes from the web into their platform. and it's a daily discovery on how this can improve their endpoints. And we are seeing a lot of tractions from the category of agentic workflows where the important part is the agent needs to take an action at the end and that's what's really measured. measured. It's really important to have the AI model take the right action and make sure that it doesn't require human supervision.

15:27And I personally think that this is AGI. It's when a model can operate independently from the user that is supposed to supervise it.

15:39Antonio Mallia:Yeah, and it's really interesting. And I guess, yeah, you've come a long way very, very quickly. What does the next six or 12 months look like? Yeah, I think the next, you know, six, 12 months, well, pretty much like, we put together a product very quickly, that's true. But we personally think this is still a product that needs to evolve together with our customers. So in the next six months, my personal goal is to scale to different use cases and see where these use cases need, where they need the web search, where they need external context that comes from the web. And as I said, I want to come up with new retrieval methodologies.

16:28I think this is one of my goals, to adapt the way we think about retrieval, we think about search, to the way these agents use the web. One example is, right now, there is a common misconception that an AI agent needs to run one web search query. And I think this concept comes from the fact that web search, in particular, when you wrap another search engine on one of the traditional ones and so on, because it does this multi-step approach where there is a layer that goes and calls an external search engine, kind of like a traditional one, and has to scrape the page and so on. It's kind of like very slow and very heavy and also very expensive at the end of the day.

17:22There is this misconception that you need to run one query and, you know, get this top 10 blue links, maybe like some snippets and augment your model with that. You know, we challenge that. We let our customers run hundreds of queries in parallel. They can literally run in parallel, get thousands and thousands of documents back, thousands of highlights, and use that as a context. It's really the yellow lemon that needs to pick which document is important, which document needs to be cited, what kind of information is required. So I haven't seen in the market yet that paradigm shift where agents run hundreds of queries in parallel.

18:08And we are allowed to do that. And this is kind of like the only way to see one concept from different angles. Just ask so many questions in parallel and get around that single topic so many different perspectives that will bring up the right context for you.

18:27Antonio Mallia:Yeah, I think that's what I need. You know, I've got agents that are constantly querying for European tech stories, right, across. They're constantly, like, Googling, searching, like, yeah, and it's, yeah, this sounds perfect. And only if you own the stack, in a way, you can customize the engine, the deployment that you have for a customer to a point where they can run hundreds of queries in parallel. they can run multi-op queries that depend on the previous one. So, you know, keep sort of like a memory of the session that they are running. And, you know, we let them do that and the customers are allowing it.

19:08Antonio Mallia:Yeah, that's amazing. I need to get on board. Well, look, Antonio, congratulations on the round. Thank you for the time. It's super exciting and I'm really excited to see your journey. You are flying the flag for Italy and Europe out in San Francisco, which is great to see. So, yeah, thank you and congratulations. thank you thank you everybody