00;00;00;09 - 00;00;25;00 Charna So. Hi. I'm sure a party and you're listening to open source Data. Today I am talking to Milosh, who's CEO and co-founder of Deep Set, an AI based company that empowers developers to make NLP part of every application and every enterprise with their platform. They've also built the haystack, the open source framework that makes it possible for every developer to power software and products with state of the art NLP. 00;00;25;02 - 00;01;02;29 Charna So Milosh, though, is renowned for his significant contributions in the field of NLP and AI. I mean, he's consistently been pioneering force in the industry and his innovative approach to A.I. and and IP has revolutionized data search processes and brought about a new era of user friendly and efficient enterprise search systems. So and while I knew ahead of time we had some things in common, we found a lot more common ground from predictive maintenance to time series and signal processing into joining or founding a startup and an LP around the same time period in 2015 and 2016. 00;01;03;01 - 00;01;40;21 Charna And now building infrastructure to solve some of those problems in the space. We talk about Haystack, Deep Set Cloud and the future of the technology, as well as a great deep dive into trust. Listen to the end to hear Milosh, his advice on getting into this space and what question he wishes more people would ask him. With all of that, let's welcome Milos. 00;01;40;23 - 00;01;42;05 Charna So happy to have you on the show. 00;01;42;08 - 00;01;45;12 Charna Thanks and flattered. Thank you. 00;01;45;15 - 00;02;14;07 Charna Are you and your background is so varied, right? I, I also this kind of person where I've worked in many different industries, whether it's defense or startups and things. And when I was reading about, you know, your past before Deep sat, it was you spanned energy cooperatives and microgrids. And so I guess my first question is like, tell us more about your journey and how did it lead to the founding of Deep Set? 00;02;14;10 - 00;02;40;05 Charna Sure. I happy to do that. I think so. You're completely right. Right. So I there was a lot of stuff in energy systems, and later on I work at predictive maintenance for the rail industry. I think the common theme is always that it was in the space off. You know, Max or machine learning. So I was always passionate as a child about mathematics, about formulas. 00;02;40;07 - 00;03;00;08 Charna You know, when I went to university, I was working early on for an application, so I tried not to be too much on the theoretical side of things, but rather look into applications that excite me. Yeah, and this is how I actually, you know, walk through all of these applications. Some of them I did at university back then in Munich and in Berkeley later on and work. 00;03;00;08 - 00;03;25;04 Charna Right. I work in predictive maintenance for the rail industry. It's systems, all of that. But the common theme is always, you know, formulas and models of, yeah, well how I ended up with natural language processing and large language models is mostly because of my co-founder myself, who like, like same as me, was always passionate about formulas and models. 00;03;25;04 - 00;03;49;19 Charna So we had some of this shared background that is shared passion, but that was the one who work already on natural language processing systems in 2016 and 2017. And it when, you know, that was the wave when NLP was very early or compared to today's capable, it's very early and very weak in capabilities, but still it was used and applied to use cases like recommendation engines, right publishing or an online advertisement. 00;03;49;22 - 00;04;16;23 Charna And this is where he got a lot of exposure into natural language processing and he excited me about using mathematical formulas. Right. Or models for understanding human language. And look, we were, I think like we didn't question use cases too much. Yeah. Know where we thought about what to do because it was somehow obvious to us that, you know, computers and machines understanding human language, that's like endless opportunity. 00;04;16;26 - 00;04;44;15 Charna We cared more about how to, you know, what, what does a product look like that gets adopted or like that helps people to adopt these capabilities and, you know, like ship them to use cases. And this was the motivation to build deep set. The challenge when we started beachhead was that we didn't have a clue how this and innovation will be shipped at all right and how it will. 00;04;44;19 - 00;05;08;13 Charna Yeah way to serve multi multiple use cases and this is why we actually started to, you know, go out and sell our understanding and our knowledge and natural language processing. And we built custom software for enterprises in order to somehow understand, you know, what's the common denominator, how do solutions look like, you know, that really solve problems in the industry and in enterprises. 00;05;08;16 - 00;05;15;17 Charna And from that exposure and from the data we gathered from it, that journey, this is how we then in the end came up with all products. 00;05;15;19 - 00;05;31;26 Charna Very cool. And so I guess there's been an evolution, right? Because you, you built Haystack, but now you're building deep set I cloud or I forget how you're calling it but sort of What's that? 00;05;31;26 - 00;05;32;20 Charna It's a cloud. 00;05;32;26 - 00;05;48;08 Charna Yes. What's that. How did you make that evolution? You went from, you know, let's discover what the market needs Then you founded Aha. And now you have this offering. So what's the concept and the benefits behind the idea? 00;05;48;11 - 00;06;27;15 Charna Hmm. I mean, you know, both like we see it as one product in the end, right? Heaps of call. This is in the and you know, it contains haystack and the whole product and all consisting of haystack and of the additional components and each so called is all build out of our experience and all of you know I always say dock food before you had any dog food you know so we we we build and solve problems and then you know you look what would have helped us maybe around haystack was pretty much to, you know, gifts developers who are very clear about the use case. 00;06;27;15 - 00;06;50;00 Charna They want to solve for it to give them to give them all components to really build a full stack solution API to end points or send data to an aquarium in between, you know, many steps that are required because as you are aware, it's not just one single model. It does the job of, I don't know, question based or semantic search, whatever it is. 00;06;50;00 - 00;07;09;20 Charna You know, it's actually it's a it's a sequence of models and other components that you need. So that was the idea about Haystack and we thought initially that if you solved the tech stack problem so, you know, this text is separated and people don't have all components in one place and in one standard. Yeah. And we thought, you know, we're solving it within a haystack. 00;07;09;22 - 00;07;30;12 Charna Then people have all in place and people will start to build and we will see how all of this thing and all of these applications will be easily served. But the reality is that, you know, it's not just a technological problem. It's a problem around many topics, like, for example, how do I know how good my l m and LP application is? 00;07;30;14 - 00;07;52;27 Charna And can I be sure enough that I don't know? You know, it doesn't give out wrong facts or anything that's critical to my use case. So am I really safe to ship this into production? And once I am in production and do I always have control and visibility on performance, you know, on hallucinations, on truthfulness, forgetfulness, all of these things. 00;07;52;27 - 00;08;23;05 Charna And that is when tons of tooling comes into place, you need a lot of tooling, you know, many features, metrics. You have to track various things in very creative ways because the way we build Web labs is so different from the way we did machine learning in the past, right? I always say, like I come from time, serious problems and times are as if, you know, like if you're solving, if you're solving time stories, prediction, what you do is you build your model and then you benchmark it against the past time serious, you label what is your gold label. 00;08;23;05 - 00;08;35;28 Charna If you're running a summarization on, I don't know, on risk reports and financial services, Right. What's the benchmark? What is a good summary summary? What's a bad summary? This is very hard to assess. Right. 00;08;36;00 - 00;08;36;11 Charna Right. 00;08;36;15 - 00;08;55;19 Charna You need, you know, new workflows and you processes and all of these new works with the new processes. We realized this are very cumbersome. And in the end, what we need is infrastructure and tooling to properly perform those those tasks and, you know, accelerate the adoption of our lamps and, you know. 00;08;55;22 - 00;09;19;00 Charna Yeah, absolutely. I mean, I'm on the World Economic Forum, it's called Got I Governance Alliance, and we just released this paper on Responsible and Sustainable AI applications because there's three different working groups. I'm on the applied side because I'm pretty much always the applied person. And this is a big topic there, right? This is a thing that we debated about a lot. 00;09;19;00 - 00;09;59;04 Charna You know, this idea that we need to not just have a human understanding of what is good, you know, is is good, factually correct, is good grounded in the data that you provided is good based on the impact it has on society. And so, you know, as we have been talking about rag and, you know, bringing enterprises to LMS in a way that preserves privacy, etc., there's this question of groundedness or truth and how do we how do we systematize that? 00;09;59;04 - 00;10;19;21 Charna So, you know, I know that I read a blog on your site probably it was a couple of weeks ago at this point about building a trust layer. So tell me tell me more about that. How are how are you or your customers leveraging RAG and what is this trust Slayer that about that you're building right. 00;10;19;23 - 00;10;51;00 Charna This trust layer is unfortunately probably right is not or cannot be like you know really a single layer single workflow step you run through, right? It's not just a traffic light system to tell to the end it's green or it's rather then you can go or you cannot go. So cross layer means for us mostly. So on the one one side, everything you need for a proper evaluation of LMA applications or as we call them in the haystack world pipelines. 00;10;51;02 - 00;11;17;29 Charna What does this mean? This means on the one hand, we need a way to somehow quantify groundedness, right? So, yeah, I don't know responses that we're seeing. How likely are they really? You know, in my knowledge base, in my database, how likely is that? They are really in there. Right. This is a very quantitative way. But the other view is more like a qualitative aspect, you know, qualitative perspective. 00;11;18;01 - 00;11;42;04 Charna Do my users like what's coming out of this and, you know, do they like the representation? I give you one idea. So I mean, RAC in general has this great you know, this great kind of buy it on, I'll call it actually like characteristic that usually you should, you should be able to so you get an answer, you ask a question. 00;11;42;04 - 00;12;02;12 Charna I don't I'm like what is what's the what's the what's a good recipe for carrot cake? And then you get a long response. Yeah. And then you get documents that were filled and all that were retrieved. And look, that comes out of this response. And even though this feels for a tech technical person like me somehow sufficient, like, yeah, it's a tick box, you know, there are the documents. 00;12;02;15 - 00;12;24;23 Charna If you want, you can check it still. You can see that users don't trust it enough, right? So ideally you can somehow really put a footnote behind behind each fact in your response. Right? That really links back to a document. And ideally you can not just reference to a document that has 2100 a thousand pages, but you actually you know, reference back to the actual fact and a passage where was this retrieved from? 00;12;24;23 - 00;12;44;00 Charna Right. And and, you know, you have to test this out. You know, you have to give it into the hands of people and people have to, you know, have the ability to try out different things and, you know, to also give you a thumbs up and a thumbs down. And based on that feedback, this is how you improve in the end, your product in your application. 00;12;44;00 - 00;12;59;16 Charna But this is also part of the trust layer. I think the biggest risk with new technologies is that they work very well, that you can somehow build with them and then you ship with them. But actually this is not really how people you know, that's not the experience people want to have. This is not what they want to do. 00;12;59;17 - 00;13;33;16 Charna And this is not what they trust. They want to have more understanding about what's behind. So this is also for us one part of the trust layer. And this is not, again, a very technological product related response. But yeah, if you think if you think about it from a cultural standpoint or like, say, a change management standpoint, we also think that the development process, the line pipelines should include end users or, you know, representative end users, people who are not engineers within the whole development process. 00;13;33;16 - 00;13;59;06 Charna Right. And when we talk about you get gather some qualitative feedback, you see what people like and what people trust. That means you are involving people who are not engineers, so not product managers. You in most probably end users. Right. And we think this is something that again, creates organizational trust in the end. You say as a CIO or even as a CEO, look, we ship this application, it's available to all of our end users. 00;13;59;06 - 00;14;22;23 Charna The chat bot on our website. But you know what? Just bright engineers who are part of the solution building, but also people who actually use this on a day to day basis. Right. And we orchestrated all of those opinions and all of those skills in one process. And this is why we can sleep very well with this in production, because we think, you know, it has been built and tested and evaluated from all sides. 00;14;22;23 - 00;14;33;09 Charna So this is for us the trust, quantitative, a quantitative and qualitative evaluation and involvement of different stakeholder us. And yeah, into this development process. 00;14;33;11 - 00;15;00;23 Charna Yeah, absolutely feels like in order to make some of these applications, especially with all I'm successful, we're going to have to include sort of a shared responsibility model where it it isn't just that the person building the application needs to understand where is their data coming from, how is the application behaving, what's you know, the impact on the user. 00;15;00;26 - 00;15;25;03 Charna But also there's some level of if we don't involve that user, if there's no transparency, well then there can be no trust. You know, I think of this as a more human level problem than anything, right? Like, that's part of what the promise of natural language interfaces is. It's like we're trying to have a humanistic interaction and and for this human interaction, we have to like, build trust. 00;15;25;03 - 00;15;54;11 Charna And that isn't a cliff. That is a series of interactions in dialog to build this trust with people. So absolutely, I agree with you on both the quantitative, qualitative, that side of things. So I have heard that you recently secured a funding round. So that brings me to thinking, you know, what's in the future, What are how are you thinking about evolving, boosting the innovation operations? 00;15;54;11 - 00;16;00;05 Charna What's what is the plan there? 00;16;00;07 - 00;16;22;06 Charna I mean, the key for us is, of course, you know, to to help the market or, you know, to go to market now with the tools we have and not on a broader scale. Right. So this means that this investment is, you know, for us mostly also an investment into, you know, the resources that are required to, you know, to bring our product to the market. 00;16;22;09 - 00;16;31;01 Charna And this is definitely that's definitely one, you know, one big chunk of investment that we're focusing on right now. 00;16;31;03 - 00;16;52;24 Charna In parallel to this, of course, you know, it's a fast moving space. Right. And it feels like, you know, this whole surface of the space increased so much over the last one, one and a half years to a degree that, you know, you really feel like, well, I mean, we always felt we're you know, we are very early. 00;16;52;24 - 00;17;16;20 Charna We were early in the market and we always were somehow ahead of the curve. And even though we were constantly in that situation, we never lost, you know, the paranoia, if you will. Right. That it's can go very fast. And tomorrow you're like, you you're not ahead of things anymore or on top of things. Right. And of course, for us, it's very important to continue to to innovate the product side. 00;17;16;23 - 00;17;43;19 Charna And for example, multi modality is a big problem because what we are seeing in enterprise adoption is pretty much that, you know, text data is one important source, but it's definitely not the single source. And many use cases usually require a combination. Of course, images, video as one thing. But when we think about enterprise use cases, we mostly see, for example, I don't know, tabular data is data, all of these things. 00;17;43;19 - 00;18;16;12 Charna So that is definitely one very, very important angle that we want to take. And then we're also still looking, of course, on on agents, right? Because when we talk about Rack, we talk mostly about use cases that we would, you know, call information related information accessibility use cases, right? Because if you use a summarization pipeline or if you use question answering or whatever you use, in the end what you do is you take all this information and you transform it, right, and make it more accessible. 00;18;16;12 - 00;18;43;27 Charna And we present it in a way that it's easier to digest for an end user. The promise of the agents is pretty much that we can not just use it to you know, get information out of knowledge basis, but to actually optimize or automate many processes, right. And automate workflows and you know that there are more and like there are more cumbersome things in the life of a human than just, you know, finding the right information. 00;18;43;27 - 00;19;04;07 Charna It's about how let's find you know, let's find let's find a slot for dinner with, you know, five friends, right? And let's sing all calendars and reserve a spot and a restaurant that everyone likes, right? This is a that that's a cumbersome process, right? It takes a lot of mental capacity from humans. But I think these things, you know, there are plenty of goals and enterprise set ups. 00;19;04;07 - 00;19;24;27 Charna And in the business world as well, where agents, you know, really, you know, can deliver on. And I think also for us, it's still the case that, you know, for us it's important to continue to invest into into exploring agents, into exploring the architectures, how they work, how can we make them trustful or like, you know, how can we increase trust to transform transfer? 00;19;25;00 - 00;19;50;18 Charna BARANSKI What can we measure about them? How can we improve them? And because, of course, it's still somehow early and everyone is aware about the challenges with agents and how reliable are those. But I think this is definitely also an area where we are strong supporters of and we want to continue to invest in there in general like make make the adoption I also possible and yeah practical to to adopt them now. 00;19;50;21 - 00;20;21;07 Charna Yeah absolutely we're seeing similar I, I, I work at data stocks. I lead the operations for what we call ragstock and we work with customers on building their stack for ag. And time and time again. Right. You see the evolution. It goes from summarization for use internally to Q&A to chat bot but there's this, there's this barrier to some companies to some applications of can we get to agency. 00;20;21;10 - 00;20;58;07 Charna There's a, there's a risk component of it where you know, what is the risk versus reward. What's our risk of doing this? What's our risk of not doing this? Can we afford not to do it? And so it feels like we're at this tipping point where the agents are starting to be considered for real industry. You know, a lot of my early conversations had to do with how can we make sure we constrain this as much as possible and we don't just have an open text box to our end users, like we want to prevent all the kinds of things. 00;20;58;09 - 00;21;19;00 Charna And now we're we're towards the point of, oh, if we actually want to get to transformation, we have to do some of this agent based framework where they can make decisions and automations and etc., etc.. So yeah, we're seeing we're seeing the same thing. Okay. So we talked a little bit about the app side, a little bit about Haystack. 00;21;19;02 - 00;21;45;20 Charna I want to talk a little bit more about you and your perspective. Right. You've got such a varied history and all of this expertise. You know, I have usually two or three questions that we go through about your experience. So feel free to share any personal anecdotes or stories during this. This part of it, of course. But I guess from your perspective both, you know, what you've seen over time and how this the space is evolving. 00;21;45;23 - 00;22;01;10 Charna What does open source data mean to you? You know, all of our guests tend to have a different answer to this question, but it's usually a personal about like your how you come about. The problem. 00;22;01;12 - 00;22;32;11 Charna First of all, like a lot right. I think it's it's a lot. But if you ask me for the one thing that's like the first thing that comes to my mind, right when I think about open source, Open source data. Yeah. Then it is trust because I, you know, I'm, I'm tend to use sometimes a very maybe abstract, maybe also not really true example, but this is how I think about it. 00;22;32;11 - 00;23;03;11 Charna So if I look at databases in general, like let's not talk about machine learning or transformer models because you know, you can you could argue it just because they're open source, they're you know, you can trust them more because, you know, in some parts we don't really understand what's going on. Right. But let's take a database, right? So if I think about a database and every database is open source and I look at that and, you know, now I am a company and I want to I want to start to, you know, to on to organize my data, have to pick a database. 00;23;03;14 - 00;23;25;07 Charna And I want to be sure that when I approach my data, I would say it's time series data. There isn't a bulk or anything in this database that automatically adds a plus one to each data point, right? So that's not what I want. Yeah, I want to be sure that I understand how does my data flow and what can happen when I do this? 00;23;25;09 - 00;23;48;09 Charna How does it get query, What's the logic behind it? And is it really you know, is it really fit for the things I want to do with my data? Like really, really and, and not because it's somewhere in a sales deck or whatever, right? Or in a spec sheet. So I want to have and I want to be able to look into it. 00;23;48;09 - 00;24;10;13 Charna And now I think even that just because people are adopting open source doesn't mean that they're always looking into it right before they're also buying a ecommerce solution. But just that aspect of community, you know, just that, just this knowledge about, hey, there are people who contribute, there are people writing papers about it, writing blog posts, writing content, sharing what they build with it. 00;24;10;15 - 00;24;45;16 Charna There are communities around it. I think this is a very big trust layer to just take this term again, that and, you know, I think for infrastructure, when people just want to adopt like plan infrastructure, right, and think about it like more in a platform sense about like what do I want to build with these infrastructure pieces, then I think this is the big value proposition of open source and that, you know, creates trust for communities that it's open, that I can theoretically look into it whenever I want. 00;24;45;16 - 00;24;46;26 Charna I think that's the key. 00;24;46;28 - 00;25;18;19 Charna Yeah, Yeah, absolutely. I, I wonder, you know, if in the past the way I've kind of seen open source is typically the builders of the project are often representative of the end user. And it's interesting because in this space that's may or may not be true, right? We're building infrastructure around components that we want maybe average developers to be able to use. 00;25;18;19 - 00;25;52;07 Charna So so I'm not sure I would love for the trust and transparency to be there and for there for an average developer to have the confidence to be like, All right, if I looked into this model, I could understand it. I wonder how it's going to evolve, right? Because if if you don't know how to build that model, you don't know how to build that transformer or you don't know how to inspect that transformer, do we need to start making the lessons explain themselves right in a way that folks can can better understand? 00;25;52;07 - 00;25;57;23 Charna So sort of rhetorical I don't have an answer to that question. 00;25;57;25 - 00;26;32;17 Charna And no, I understand where you're going it right. And I think, look, that was my point. That's why I picked the database example, right? Yes. You know, like, of course, transformer models and alarms have different flaws. Right. We're all aware about. But I think I think so. This is now maybe like, you know, I'll tell you with a grain of salt and it might sound a bit you know, don't get me wrong, the way I phrase it right now, I'm also not a native English speaker, so this is why. 00;26;32;20 - 00;27;04;06 Charna But I think, you know, it's I think the illusion of transparency is often enough. Right? So the illusion or actually the, you know, I have the opportunity if I really feel I want to do this and I really want to do it again, I have the opportunity to do it myself. But in the meantime, what I see as a living repository, people talking about the stuff, praising and telling me they build with that. 00;27;04;09 - 00;27;27;20 Charna And, you know, it's trust that on again, quantitatively and qualitatively, having good reviews, I'll look at the number of contributors in the repository. I'll look at different stuff. I feel this is a life. There are people who work every day with it and you know, they share very openly. Any feedback to the code Banks, to the whole project, file with me and all. 00;27;27;20 - 00;27;50;02 Charna And yeah, of course we don't understand Transformers, but hey, there's a community of, I don't know, 100,000 people, 1 million people, whatever was building with this, like, you know, that's fine. But this is, it's okay, right? And this is how I pretty much think about it. So I think that makes sense. I think explainable is is good and it's something we should strive for because it would actually help us to control models better. 00;27;50;02 - 00;28;10;00 Charna Right. If we understand really what's going on and why do models hallucinate? You know, this gives us the opportunity to, you know, effectively develop strategies to mitigate these problems. So this is why I absolutely in favor of it. But I think it's not critical to the to the adoption, to the mainstream adoption. Yeah. 00;28;10;02 - 00;28;34;13 Charna Yeah. Very, very valid point. Like the living, breathing, breathing community aspect certainly brings a lot to the table. I know that, you know, even with building a company in and in, you're doing this right now, part of what you have to do is hand off some level of responsibility and autonomy to the people that are building things. And you can do everything yourself, right? 00;28;34;15 - 00;28;51;22 Charna So I think that's a yeah, it's a great component that will always be there of open source is the ability to to look into that community and understand, you know, what are people's concerns. Even if, you know, you don't even know, you should have that concern. Very cool. Okay. 00;28;51;27 - 00;28;58;10 Charna Yeah. It's interesting. You don't you don't have to ask the community's about about their concerns. So they don't share. 00;28;58;13 - 00;29;04;26 Charna Every every discord I happened to. It's just right there. Great interface. I love it. 00;29;04;28 - 00;29;12;14 Charna So I think the fact there's, you know, that's a very effective way. So I think it's very effective now that people are like this. I think it's very good. 00;29;12;16 - 00;29;43;17 Charna Yeah. Yeah. So we're coming towards the end of our time. I have just two topics left. I guess I'm very curious to know what's the question that you've always wanted to be asked, but no one has ever asked you something that you should have been interviewed for on any topic, whether it's work related, unrelated to work. There's something that's important to you that people should know and no one's ever asked you mean. 00;29;43;20 - 00;30;04;10 Charna But this is a very thought provoking question, by the way. And like I might I might misinterpret it a little bit because I think it's a bit like, what am I proud of right? Where people are not acknowledging properly, you know, Oh, maybe one. 00;30;04;12 - 00;30;04;29 Charna That's one way. 00;30;05;06 - 00;30;06;25 Charna I know this is fair. 00;30;07;02 - 00;30;28;26 Charna This is that's one when we I've had people interpret another is a lot of founders for example think that there are topics that we just don't talk about that we should talk about like what's hard you know, isolation and burnout, things like that. So there are two sides of a coin, though. 00;30;28;28 - 00;30;57;04 Charna Now. So I mean, I think actually what I would and this is not about acknowledgment and not but I think what I would like to talk about more, what I what I what I would like to be asked is like, how is it to be really, really early in the very early, early market because this is what we what deep said was right. 00;30;57;04 - 00;31;24;22 Charna I mean, people are talking today about an early a market or early low end market that is just, you know, just has been born. And for us, it's very interesting to look into it because it feels like, well, yeah, for us this is a little bit like the status quo of the past six years. Yeah, right. So I think I think and this is not you know, this is really like not about I acknowledge that I'm so early. 00;31;24;22 - 00;31;47;28 Charna It's more about I think it it's useful, you know, for people to understand a little bit these early market challenges and you know, to be curious about also understanding what we think, how we can, you know, navigate those times. Right. And what what is important. And because to be honest, it didn't change too much from 2019 to today. 00;31;48;00 - 00;32;01;24 Charna Yeah. If I think about what is needed but I think this is this is probably a question I will I would like to be ask and maybe to some people ask it also like in other variations, but I just put it here. 00;32;01;27 - 00;32;35;01 Charna So that's great. Yeah, certainly. I mean, I go back to was it 2015 was the first an LP start up I had worked for. And you're right, we had many of the same means. It feels to me part of the challenge of being early. And we'd love to hear if this is your experience as well. It's very hard to kind of motivate the need for some of these things when not everyone is paying down your door. 00;32;35;04 - 00;33;04;25 Charna You know, we were at Texto, we were building a writing product that was supposed to tell you how your words would perform, not necessarily would someone like it, but would would you get qualified people to apply for a job? What would be the distribution of, you know, gender or various components, like what's the outcome of this? And the tooling just wasn't there and it was hard to be, you know, you're a user or you're a builder and you're like, Build me this thing. 00;33;04;28 - 00;33;09;29 Charna And the community is like, Well, nobody needs it. So for you. 00;33;10;01 - 00;33;42;13 Charna This is you know, this is this is definitely one important. And I think so it all comes like, you know, being in early market all comes down to the use case and not just the use case, but the business case behind. Yeah, why what is the motivation to surf this and fulfill this use case? And even if you end up and this is, by the way, possible and and this is I think what we are seeing right now, to be fair, right, is many enthusiasts, but people who are enthusiastic about the technology per se. 00;33;42;13 - 00;34;02;28 Charna Right. I love the tech. You know, I love to spend time with the technology. I think you know what I'm talking about. I'm saying, you know, like we can you can you can run and like we we could chat for this for another hour about this. So like, you know, this this is that's nice. That's fair. The problem is maybe, you know, maybe you get the interest. 00;34;02;28 - 00;34;23;05 Charna Maybe you earn some bucks by working with them or selling them something. But as long as there isn't in the end, you know, a compelling reason why this should really, really find its way into a productive application. But there is no way for you to make, you know, like to to earn and to have a business in the in the mid and long term. 00;34;23;05 - 00;34;43;11 Charna And maybe I talk a lot about this entrepreneurial perspective right now. Right. But I'm of course, you know, all of us in the space, like entrepreneurs, startups, corporates, academics, it doesn't matter. All of us care about one thing. How does this actually create value, you know, and how will this actually get adopted and how did this move into production? 00;34;43;13 - 00;35;13;05 Charna And I think the answer is it's not enough to be excited about the technology and it's not enough to just give the technology into people's hands and think that, you know, they can come up with something. Right. I think it's a lot about being close to what is this used for where you know, where it's value and then probably optimizing what you what you can offer pretty much for that and make it very, very easy for organizations to use it exactly for these high value use cases in order to succeed. 00;35;13;05 - 00;35;31;19 Charna And then, you know, then fantasy and creativity is happening and people have plenty of ideas and you will see everyone wants to see this. And each and every application, you know, this is this is great. But I think in an early market, you have to care a lot about what is the end and and call for an end user with what is built here. 00;35;31;19 - 00;35;35;24 Charna Right. And selling technology just for the sake of technology is not possible. 00;35;35;27 - 00;36;01;20 Charna Yeah no it's absolutely not possible. I mean, I forget the last numbers, but it's something like ten times more expensive to do something with Malala than it is without. And it's it has to be worth it, right? It can't just be solve the same problem with a different model. Right. If we have to have the right impact and we have motivate, this is the reason we should go this way. 00;36;01;22 - 00;36;47;21 Charna Not to mention the change management required to make people adopt a new technology. Right? It's just drop it in and you go, it's upskilling, training, the whole thing. The whole thing. All right. So the last question, we're almost there. So as your journey has gone through this early mark and as a founder, as a CEO, creating something new that just wasn't there before, I know our listeners are always super curious to know what it vise, whether it's something you have for them, either in stepping into the world of an LP that's moving as fast as what deep set is, or, you know, how can they get more miles per gallon out of their day to 00;36;47;21 - 00;36;57;01 Charna day effort to go further and faster in their journey with data and alarms? What with advice might you offer? 00;36;57;04 - 00;37;23;28 Charna My advice as well. Focus. And that means for me, you ask like you have to spend a lot of time with something. Well, in order to really, you know, in order to really be able to succeed, I will like and let's say, outperform others with it. Right. So it's a lot about investing a lot of time and being very consistent. 00;37;24;00 - 00;37;51;12 Charna And this and this implies also patience and stamina. So you hardly can expect that, you know, within one week like, oh, I'm great and I'm now the new pro and I can do everything and I ship production. I know what's going on, right? So I think, you know, people who want to enter this space and who really want to make a career and it should really be aware that it's you know, there's a lot of surface, a lot of surface. 00;37;51;12 - 00;38;12;03 Charna And it just grew exponentially within the last 12, 18 months. Right. And it will take a lot of time and yeah, it will take a lot of time. And the best thing to get there fast, if it's it takes a lot of times to invest a lot of your time into it, right. I mean, you should be very focused on these things and understand, you know, how those things work. 00;38;12;03 - 00;38;33;19 Charna What is rag, what are mitigation strategies, what is a hallucination, really? What does that mean? Where does that come from? I've got to like which strategies kind of follow up? What does it mean to move to production? What what does this term production mean? Right. And again, all of these things take time. And the faster you want to beat, the more time you have to dedicate during a day into it. 00;38;33;19 - 00;38;54;04 Charna And I think this is where my advice when we started this, that we were forced to do it. We didn't we didn't have any choice, right, to care for anything else. So that was good. That that was that was important. That really, you know, disciplined us. But I think my advice is and don't try to rush things in this based on don't think don't think. 00;38;54;04 - 00;39;21;28 Charna Also, no one is an expert and one week maybe not even one year, you know, but it's worth investing your time. But do it and be very, very focused. Invest enough time and have respect. It's a it's a tough discipline. A lot of surface, many bright, super bright people are right now shaping the space, are very impressive teams out there in startups and research institutions and, you know, also big organizations, Fortune 500. 00;39;21;28 - 00;39;46;01 Charna I'm really impressed when I talk to people and see what kind of people are out there and all of them are contributing to this. So, you know, it takes some time. Be focused on that. Now. Don't rush things. Yeah. At the same time, for companies, that also means you cannot rush things. You have to be very focused on, you know, what's value, where is it, where can I find it, How can I optimize for it? 00;39;46;03 - 00;39;50;17 Charna That's very important also in startup to not be all over the place. 00;39;50;19 - 00;40;18;11 Charna Absolutely. I've I've been in start a play for most of my life and I it's hard for me to imagine working at a large company about even so it right now the excitement from startups from large companies. It's it's there and it's very hard to be patient and slow down and focus but it's going to be necessary right Like there are a lot of problems in this space and I think. 00;40;18;13 - 00;40;22;02 Charna It's that being patient doesn't mean that you're. 00;40;22;04 - 00;40;40;25 Charna Yeah, it doesn't mean standing slow. It's standing still like you don't have to stop. But being patient with the outcome, you know, how do we get there? How do we do it the right way? You can push really hard and put a lot of energy into it. But yeah, you know, we're working towards this future, but so wonderful. 00;40;40;28 - 00;40;45;02 Charna I think. I think like a you know, you. 00;40;45;05 - 00;40;45;27 Charna Can continue. 00;40;46;01 - 00;40;47;10 Charna To have something there, but. 00;40;47;12 - 00;40;48;09 Charna No, go for it. Go for. 00;40;48;09 - 00;41;13;18 Charna It. I just want to add one thing. So I think just because you're patient, it doesn't mean that no urgency. Right. And I think what's important is learning to learn the way machine learning models learn. And they learn by failure, Right. And by errors they're doing. I think I think the worst thing is, you know, to procrastinate like shipping something, right? 00;41;13;19 - 00;41;28;20 Charna You should always be shipping something that is or makes them an expert in end how you do it. You observe it or what is happening, how do people perceive it? You iterate over it. I think this would be really like it's not about reading papers the whole day, right? It's about, you know, building, getting your hands on it. 00;41;28;22 - 00;41;45;24 Charna But again, to have a perfect system, to have a use case of, you know, for 500 million users, whatever, you know, high level of large scale, high value, I think that's going to take some time and it's just not going to be an overnight success. 00;41;45;26 - 00;42;11;13 Charna Yeah, Yeah, absolutely. We tend to say opinions are inside, facts are outside. So if you're not shipping something to get what people actually think, you haven't really made it progress. Yeah. Yeah. Well, thank you so much for this wonderful conversation. I am so excited to see where you are and Deep Set is going into the future hopefully, you know, maybe maybe a year from now or just expense for now. 00;42;11;13 - 00;42;24;03 Charna We can check in and see, you know, how things have been going. But thank you so much for being on the podcast and I know our listeners are going to love love your your, your episode. 00;42;24;06 - 00;42;25;05 Charna Thanks a lot. Nothing. 00;42;25;06 - 00;42;55;03 Charna I'm backstage with our executive producer Leo Godoy to cover his takeaways from the conversation. Leo I thought the conversation with Milo was incredibly information dense. It's not really a surprise from somebody who's been working in this place for so long, even in the early market. And I know I'm going to go back and look into how they're handling trust in the A.I. trust layer from a quantitative and qualitative perspective, like deep in the code, I really want to see how they're handling that, but I'm interested. 00;42;55;03 - 00;42;56;13 Charna What stood out to you? 00;42;56;16 - 00;43;32;22 Milos Well, Cerna I think to me, you guys sharing this common ground of experiences, you know, like really with some particular happening in both of your careers, I think it's such a funny thing to have on the show in a good way because you guys know how to approach the technical aspects of it in a way that it even, you know, with all this different types of guests that we have around, it's just so hard to find because you can connect on different sets of, you know, knowledge and experiences. 00;43;32;22 - 00;43;57;01 Milos So to me, it's so important to have Milosh around because he's able to share with you a lot of a lot of his life, you know, even in a personal level that, you know, other guests will have it in different ways. And also when he's explaining everything that open source means to him, I think it's just something that you can relate very well, right? 00;43;57;04 - 00;44;29;02 Charna Yeah, absolutely. It's kind of fascinating because I'm so used to, at least in the Pacific Northwest, where I've been in the startup tech sector, there's a lot of folks who don't have the very shit across, you know, from the energy perspective, the hardware perspective, the time period, serious perspective. Unless you're talking to someone who's doing like, like with Amazon and Alexa or Apple and some of the hardware crossover. 00;44;29;02 - 00;44;45;21 Charna So it's really great to have someone, you know, all the way maybe halfway around the world has shared some of the same experiences. It's it's great. An interesting shorthand where you can just come at the problem and be speaking the same language. 00;44;45;28 - 00;45;00;04 Milos Yeah, absolutely. And the way that you guys approach, you know, the technology and the way that you approach your questions as well to him, I think it's so important to enrich the conversation in a more general way as well. 00;45;00;07 - 00;45;23;26 Charna Yeah, absolutely. Well, thank you so much for spending the time and thanks to our audience for listening. If you like the show, please subscribe and give us a five star rating on your favorite podcast platform and special thanks to the Greystoke Network's team, our producer Leo Godoy, Gustavo, our editor and Audio and Visual engineer, and thank you to data sacks for sponsoring the podcast. 00;45;23;28 - 00;45;27;25 Charna Thanks again for listening. Catch you on the next episode of Open Source Data.