AI in Reality Fireside Chat: Enterprise AI & Open‑Source Innovation

Speaker 1 [00:07]

Welcome everyone here for joining us for the fireside chat. In this fireside chat, we want to discuss AI in reality. So we want to skip all the hype stuff and vision stuff and say, okay, we are a practitioners conference. We do open source and data and AI in practice for more than 10 years. And I'm really happy to have all of you here on stage to tell us more about yourself, how you think, how you work with AI in real life. And yeah, let's start with a short intro. Walid, what's your journey? Where did you start? What's the recipe to become the CEO of Merck?

Speaker 2 [00:51]

Happy to do so.

Speaker 1 [00:51]

Happy to do so.

Speaker 2 [00:52]

So how many hours do we have?

Speaker 1 [00:53]

have uh we have all the time we need if we tell we need it okay so yeah now make it of course

Speaker 2 [00:58]

I'll make it, of course, you know, I'm joking.

Speaker 1 [00:59]

If it's short, we have some more.

Speaker 2 [01:01]

Okay. I'll try to give you short and sweet. So my name is Walid. I'm the Chief Data and AI Officer at Merck, which is the keystone sponsor for PyCon. Glad to have you here, and also happy to. And also located in Darmstadt, we are a 357-year-old company in the meantime, still family-owned in the 13th generation. I always like to joke like we are the oldest scale up in the world because if merck has one ability it is pivoting so it started off as a pharmacy it went into pharmaceuticals into life sciences into consumer electronics and in the performance materials and electronic space afterwards so if one thing is in the dna of this company it is change it is pivoting where it's needed so anything new is a is welcome to the organization it's a science and technology company so people are naturally very curious about it so honestly it's a perfect place for me to be in the data ai space and our journey is now four years in with the company i started off as the chief data officer and then as always including analytics and then as ai rose to prominence and relevance also let's say broadened the scope and the topics about it i guess that have to do it for starters

Speaker 1 [02:20]

Oh, well, let's dig a little bit deeper, because I've done some research. Where did you actually start? Actually, is it true you just started programming databases and websites before you joined consulting?

Speaker 2 [02:33]

Exactly. So my background is in computer science. Actually, I'm a big fan of interpretable languages that start with P, not Python, because back in my days, end of the 90s, it was Perl.

Speaker 1 [02:33]

Exactly.

Speaker 2 [02:44]

So the Camelbook was, and we didn't have Stack Overflow. We didn't have ChairGPT or anything that was, let's say, different kind of coding back in the days. But that was my background in programming, in database administration, network administration, system administration. Then 13 years of consulting and then getting fed up with being, let's say, like a midwife, just bringing things to life but not taking responsibility. So taking over an operational role at Mercedes-Benz in Stuttgart and then at Merck four years ago.

Speaker 1 [03:16]

years ago it's not like a curious person

Speaker 2 [03:18]

It, yeah. Okay.

Speaker 1 [03:21]

Alex, how did you, what's your journey?

Speaker 3 [03:25]

My journey? Yeah, good question.

Speaker 1 [03:27]

question like how did you how do you like we know each other from bringing python to quantitative asset manager corneum so what was your journey because you were one of the drivers behind it like yeah like why python why why not just buy software where did you start yeah

Speaker 3 [03:48]

So maybe I start with my academic journey.

Speaker 1 [03:48]

So.

Speaker 3 [03:53]

So I'm a physicist and I did my PhD in economics. So how does that work? There was a professor back then who ran a quantitative hedge fund. And this is all about statistics and computer science. But you could not give a PhD in physics for that, right? So I had to live with then receiving a PhD in economics, which didn't hurt. as well but this is when i fell in love with python and computer science and also machine learning back then of course nobody was talking about generative ai i started with emacs pre-pandas pre-psychic learn so this is where my journey started still we would do machine learning neural networks especially yeah and then i had several positions consulting enterprises embw for example and my journey then took me to where i am today as a cto and partner at kvoniam asset management kvoniam is a company that most of you won't know so i'll briefly explain what it does it is a quantitative asset manager which means we digest data we apply and analytics to it also machine learning also gen ai by now and at the end what comes out are portfolios optimized for return risk and the environment so what is called esg and as you rightly said you were one of the people who supported us in this journey from moving from a really legacy and scattered tech stack so back then we had sas our c sharp a little bit of python a lot of sql and we harmonized this into Python. Yeah, so this is how we got to know each other.

Speaker 1 [05:53]

Ines, where did you start?

Speaker 2 [05:55]

how did how did

Speaker 1 [05:55]

How did Spacey happen? Did you start with programming?

Speaker 4 [05:59]

Yes, I've always actually programmed as a teenager. I mean, I think you know the story.

Speaker 1 [06:03]

I know, but we have to bring it to the audience. Not for me, the questions.

Speaker 4 [06:08]

Yeah, I know. So yeah, as a teenager, I was kind of an indoor child. Didn't have that many friends and mostly spent time at home in front of the computer. And so I discovered that in Microsoft Word, you could make websites. And so I uploaded my own website. I was like 11 and then got really into actually writing the code, having my own blog. I did like a music blog. I had a personal blog. I did all kinds of things. But I didn't actually end up doing computer science, because at the time I didn't really feel like a programmer. You know, when I thought of programmers, those were like the boys in the computer club, and I didn't really, you know, I was like, I was a woman, I was doing websites, I loved design, loved writing, and so, yeah, I did communication science and media science and linguistics. So that was, you know, that actually ended up becoming much more relevant later on, bit of computational linguistics, worked in media for a while, and then kind of by coincidence I found this area where I could really create a job for myself and combine everything I'm passionate about, like from product to actually hands-on writing code, building tools for people to use, language, working with language, working with text. And so, yeah, I met my co-founder, Matt. He was previously in academia, left academia when he realized that companies were using his research code, which was totally not designed to be used in production. And he really saw... Also, he was at a point where he had to write grant applications. So he was like, nah, I'll focus on doing something that's really designed from day one for real products. And yeah, we started working together, founded a company together. And yeah, what we're really focusing on is building developer tools, and especially building developer tools that help companies do AI development themselves, again, in-house, take back control and really instead of just using the latest large models as the system, actually use them to create systems. So that's from helping you design models to actually training your very own fully private, accurate, fast, small models in-house.

Speaker 1 [08:10]

Actually, you could put, like, Spacey comes from messy research code. I mean, I'm just referring to the keynote. I don't know.

Speaker 4 [08:18]

I don't know how messy, Matt would take offense to that.

Speaker 1 [08:20]

to that no offense but if you attended the keynote from guara this morning i mean we were talking about space long euclid mission cosmology dark matter she said oh we had this messy research phase like also the code so actually you could also say this was a research project and you brought it into production and spacey it's really hard to find real numbers because there's so many statistic but i could say it's at least like half a billion downloads so basically it's a state of the art open source tool you brought from a research project with matthew into production

Speaker 4 [09:00]

hundreds of thousands of companies.

Speaker 1 [09:01]

like 10 around for 10 years. I think that's awesome. So what gets you going like every morning? So you already have to go one of the go-to tools for natural language processing. So So yeah, what sparked the commitment to open source?

Speaker 4 [09:19]

to open source? Yeah, I mean, I do think open source is a great way to create a lot of value for people. And I think we, very early on, we realized that there's not really an advantage in locking up proprietary software. It all comes from open research anyway. So there's not really an advantage you have by not open sourcing it. And what's much more valuable for companies are the actual use cases and the data and what you apply it to. So I think it's very, actually, it's been very inspiring to see what people build with with our software, and actually at every conference I go to, and especially here, we meet users, people at companies, they tell me what they've built, there's so many interesting use cases and so much I've learned, and it's really nice to see that the software actually can apply in all kinds of verticals.

Speaker 1 [10:03]

Curiosity and application. Yeah.

Speaker 4 [10:05]

Yeah, and the successful projects we see, those are really spearheaded by often even a single developer who identifies a need and really has subject matter expertise and says, there's a problem, I need to fix it. And it's never been so easy to get into programming, pick up Python, pick up NLP, and get going. And yeah, we've seen a lot of success stories of developers just fixing problems and doing the thing.

Speaker 1 [10:30]

Alex, which co-IDs drive AI and open source at KUON-YAML?

Speaker 4 [10:31]

Alex?

Speaker 3 [10:35]

which core values

Speaker 1 [10:36]

Yeah, why Python? Why open source? What's the reason to do this fundamental paradigm shift?

Speaker 3 [10:47]

Yeah, so we came from a world where we sat on, as I just said, like SQL and ZAS and all that. And, well, the one thing is that you always have a vendor lock-in, right, which is naturally something that you are trying to avoid at a company. And then, as you just said, so much innovation happens in open source, especially Python. And for us, as a quantitative asset manager, what we do is we work with data, we do a lot of analytics, and at the end we produce signals, right? So what do we want to trade, how does our portfolios look like? But that's, at the end of the day, that's all data and analytics, right? So Python comes as a natural choice somehow, right? there weren't big discussions around that so we had to collect a little bit in the beginning and tell the people well you cannot use R anymore, you cannot use Sass anymore, please migrate away from that and it was good and if I look where we are now with Python I think we have a great journey behind us we have reached a really professional great development state and this allows us now for the whole company to take in the new things like Gen AI and make them available throughout the whole company which I think is really good and really helpful.

Speaker 1 [12:26]

Well, what keeps you going and you have like, you have like a lot like Merck is like a really big company. We have like life science, you have electronics, which people don't know, you have pharmaceuticals. So what keeps you motivated, like getting everybody on board with data and AI and possibly also some migration and innovation and everything. I mean, it's like, sounds like a lot to handle.

Speaker 2 [12:51]

handle it is a to a certain degree yeah let's say a large large large challenge so as a company we have 63,000 employees worldwide and I always so when I started the job I had an interview with my now CEO Berlin and she asked me what is my dream and I told her my dream is that you don't need me anymore because 63,000 people in this company breathe data and AI everybody knows what it is when it's useful, when it's not useful and can apply it, then my mission is done and then I can retire. So still a few years to retirement, unfortunately. But that is what keeps me up. And I like what you said. It's when essentially subject matter expertise and technology expertise come together. Unfortunately, it is separated often in people. So my mission or the mission of my team to a certain degree obviously is to lower the entry barrier when it comes to technology. Similarly, you described did alex so we call it we give it a name we call it optimize our data and ai ecosystem that is our overall architecture and also our promise to the organization that the tech stack that we provide is secure compliant and cost efficient and more and more open source because it is it is something that that is very dear to us and we're open sourcing ourselves and the second piece is also educating and training the people so they get the skills not everybody can be a full-time developer but you get foundational skills you get low code no code environments where people can get productive and the more people on the ground know what ai can do for them and what it can't do which is equally important or maybe even more important the more they can be let's say a subject matter experts a better counterpart and ally for a developer to build the stuff that matters and that makes a difference in the business

Speaker 1 [14:43]

And I think you bring up a really important part because you don't talk how we use this.

Speaker 2 [14:47]

use this library.

Speaker 1 [14:48]

library, use this technology, basically it's all about humans, like a big part which is probably overlooked in all tech development innovation is the human migration part.

Speaker 2 [15:01]

I wouldn't call it migration in this. No, I'm sorry.

Speaker 1 [15:04]

Because it's a...

Speaker 2 [15:05]

But actually, yes, in our strategy, in our data and AI strategy, we always, I like to call it the holy trinity, and we've got three dimensions, and it's not rocket science. You can find that in multiple articles, but it's people.

Speaker 1 [15:19]

people first.

Speaker 2 [15:20]

it's the mindset and the skill set then it's about the ways of working so how do we orchestrate and integrate and the second one and the third and last one which is also important is the technology piece those are the libraries and the services and what do we use in globally what do we use in the US what do we use in China and stuff like that so

Speaker 1 [15:41]

What's your strategy? I mean very often I see people like to do what they do and stay in their routines So what's your approach like? Making them opening up say hey, there's something new we have to innovate. We have we have to we have to become better We have we have to evolve Because like very often people just say hey, I know how I'm doing this. I'm doing it for 20 years. Why should I change?

Speaker 2 [16:03]

See, that's the funny thing or the good thing about Merck as a company. So the company vision is sparking discovery, elevating humanity. So there is already a very high ambition. What we do makes a difference. And to a certain degree, it makes sense also in the pharma healthcare space where we develop drugs, where we have a large fertility business, where we help to bring babies to the world for those who otherwise wouldn't be able to do so. But sparking discovery is about the small innovations. It's about, so curiosity is, let's say, part of the DNA of the company. So to a certain degree, I just have to trigger it, but that's maybe a bit of a luxury situation.

Speaker 1 [16:42]

But also, it's value. You say, okay, we help babies. We just don't sell a consumer product or something entertainment. We say, hey, we help save lives, make people's lives better. Yeah. Oh, awesome. Awesome. Yeah, Alex. So, I mean, we started in the middle of the pandemic with migrating persons. So, what has changed then? Anything also changed? Culture? Work? like what I think probably I think like with all opening up to open source like migration always also has changed in structure and and people so what has changed since then so before people were doing doing their jobs and now we bring a lot of innovation to the company what has changed

Speaker 3 [17:29]

Yeah, very good question. We made one very important decision in the very beginning, which was to use Python throughout the whole company, right? And this opens a lot of doors, obviously. So back then, application development was C-sharp. Research, for example, was R. The two teams couldn't talk to each other. They couldn't learn from each other. They couldn't share code. They simply had to live side by side. right and this has fundamentally changed to the very good of the company obviously right so as of today the teams work much more integrated there is way more exchange which makes us faster which increases quality of course because the business people and the tech people they can work with the same language at the end of the day and for our business people it's a natural thing also to understand a bit of coding so they also understand what python means how to work with it the researchers especially they're very good in these things as well so there's a lot of harmonization happening which i think is great

Speaker 1 [18:48]

So actually, the strategy is not like individualism was not the strategy, say, oh, you like this language more, go for it. Like that means, hey, let's bring all the whole team on board. Let's normalize, let's standardize, let's improve our communication. And also, of course, asset managers, of course, many people are very skilled in math. So programming is not that far. And you basically open up and you can basically avoid a lot of misunderstandings through normalization. and bringing everybody on.

Speaker 3 [19:17]

And it makes it faster, right? So the first steps, everyone has to realign a little bit their way of working, right? That causes friction. I think we can all imagine that. But then once you are behind these initial fires that you have to live with, you really feel how this harmonization speeds you up and makes you better.

Speaker 1 [19:42]

different perspective like I think yeah ten years ago Spacey and Spacey was I would say like and immediately here and in like like was an immediate success I would say because I there was like what's it was called and I'll be the other way and I'll TK yes right yeah

Speaker 4 [19:59]

Yes, right. That was more like designed for teaching and research. At the time, there was...

Speaker 1 [20:02]

When I started, there was an LTK, and then Spacey came along, and Spacey was, wow, this is, like, so much cooler, and you basically were, like, an immediate success, like, the go-to package that makes things easier. And this, and I think that's the important part here, you also had, like, early adopters in Fortune 500, many companies, and could you give us some insight what made them go very early? We're talking like 2015, 2016 times open source is still...

Speaker 2 [20:36]

So

Speaker 1 [20:37]

I would argue for many enterprises, something exotic. Yeah, who was smarter than the others and what made them special, like cultural wise or whatever you have experienced, just like say who were your early adopters and if you have an idea of why.

Speaker 4 [20:56]

Yeah, so, I mean, 2015, that was also roughly when we kind of saw the first NLP hype and chatbots first came out. And so I think there was a lot of interest. But I think, on the other hand, we actually saw something quite similar to what we're seeing today, that, like, you know, especially decision makers, like, they hear AI, nowadays generative AI, and are like, oh, my God, we need to do something. And, yeah, some companies actually want prestige projects, and they usually fail. It's like, oh, let's just do a chatbot, and that never makes it out of the prototype phase. But, of course, there are other companies that are engaging with it in a more reasonable way. And so we definitely, we've seen that. We've also, actually, some of our earliest projects that we worked on with companies were helping teams at large organizations move their AI development back in-house. Because they had, for example, their managers went to World of Watson in Las Vegas. I don't know if you remember these TV commercials with Mad Men's Don Draper and, like, ooh, the glowing brain. And, you know, they'd excitedly get back and are like, ooh, we have, like, all of these AI platforms now. And the developers were like, okay. And, yeah, then they would, for example, pick up Spacey, run some experiments, and they realized, oh, we can build something much better and much faster, fully private in-house. We don't need these platforms that we sort of bought into. Like, what?

Speaker 1 [22:19]

a part, because I know the spaCy documentation. I think if I would vote which project has the most awesome documentation online, I would probably say, yeah, spaCy because you spend a lot of time bringing it. And so it was probably not the documentation. It's very often missing open source packages. So you can go to consulting and the consultants know how to use it because it's like a black box. So you didn't go to explain on the undocumented What did you consult on? Where were the pain points you could help them to solve? Where did they really struggle apart from how to use spaCy?

Speaker 4 [22:56]

Yeah, yes, it's funny. You mentioned actually the documentation because that was, you know, we were quite conscious about like our business model because, you know, monetizing open source is difficult. And we always saw that there was that would otherwise be this conflict of interest. If our offering was, hey, we help you use spacey, then if our documentation is too good, we don't, you know, nobody needs our help. And if we make our documentation bad, then well, nobody uses our software. So, you know, we always saw it more around like workflows and really getting getting teams productive because the hard part of machine learning even years ago was it was never the actual training um or like even developing the models it kind of doesn't actually doesn't matter that much it's really about how to take a business problem and break that down into components that you can solve with machine learning and also remembering that this is not a competition like you're allowed to make problems easier we're not in research here like i mean it sounds trivial but like actually um you know it's something i think it's very important we have to remind ourselves Like, what are the easiest components that we can build? And also, how can we build systems that work with the best practices and principles we have built up as an industry over the years?

Speaker 1 [24:04]

I can very much relate to that. I remember this one discussion with a client. He wanted to use some super fancy algorithm just like it developed and published by Uber. And it was about temperature. And I said, well, you're the physicist. I just had physics in school. But I think temperature is something we really well understand. How about trying a regression and stuff like that? Like really say, hey, we want to solve the problem. We don't get paid to pay with fancy tech.

Speaker 4 [24:32]

Yeah, exactly. And I think the companies that actually saw that and really focused on that, those were usually the projects that were also most successful.

Speaker 1 [24:33]

Yeah.

Speaker 4 [24:41]

And I think some companies realized earlier on that, hey, we can totally do our own development in-house. We can build our own systems that we own, that we fully control. And, you know, it's totally possible. Like we've learned outsourcing development never really worked. We learned that in the 90s. Outsourcing data development, annotation, we've also realized that doesn't work. We need people with a connection to the task.

Speaker 1 [25:03]

Too many layers of communication where stuff gets lost.

Speaker 4 [25:06]

Yeah, and have the teams collaborating together also across subject matter experts, machine learning developers, often that can be the same person. And I think that's definitely where most of the success happens from what we've seen.

Speaker 2 [25:21]

Early developer

Speaker 1 [25:22]

So early adapter, I read you built my GTP, you worked with OpenAI when JTP2 was a thing.

Speaker 2 [25:33]

Not me my team. Yes

Speaker 1 [25:34]

Yes, your team, you and your team, and you enabled your team. You also, and just in the other room, your team is giving a tutorial on Baby. Yes, exactly. Sorry for the bad timing.

Speaker 2 [25:45]

Yeah, I'm sorry for that too.

Speaker 1 [25:45]

Yeah, I'm sorry.

Speaker 2 [25:46]

Yeah, it's not a problem.

Speaker 1 [25:46]

Yeah, it's no problem. Yeah, yeah.

Speaker 2 [25:48]

so so but that's why the other room is full and this is

Speaker 1 [25:52]

It's more people than it looks like. Okay. Yeah, we have a capacity for 1,200 people in this room, and we have 1,500 people on site. This is really good. So, yeah, so you also now go into open source. What's the strategy? Like, what's your approach? Like, you start with open source. You're all with, you're like the best friends with open AI. You have your own GTP. So, I mean, you started early. So, there's open source. I see that you see like there's a mixture and how do you decide when to do what or probably maybe stop?

Speaker 2 [26:28]

Like, so I'm one of those executives you've been talking about to a certain degree, but I try to be not in the way when people who know stuff better than I do, let's say, call the shots or when they make suggestions. And so the fun story that you mentioned was when CheGVT came out and I saw it and I said, OK, guys, we need to talk and we need to put an NDA in place and we need to see how we can leverage that technology. And then the team just looked at me and told me, you signed that NDA nine months ago. And I said, oh, great. Well done. Good work, team. And at the time, they told me there was a research lab and I don't need to know too much about it. Long story short, what we did, or the team did in six months, was to build our own in-house clone based on the Microsoft Cognitive Services. And now then a collaboration with a startup afterwards. So we sent down the original solution and then went with a Berlin startup called LangDoc to develop our MyGPT suite. And ultimately, what we do is to balance things out in terms of what's the best for the company. If it's open source, great. that is always our, let's say, preferred go-to when there is something that is enterprise-ready that we can leverage. If it is not, then we look for partners. And partners can be big or small. If they're small, even better, because we can offer them more as a company if your partner is big. If it's a large multinational corporation like ourselves, they only want our money and our brand to a certain degree. Younger companies are more curious in terms of learning the subject matter expertise, have an environment where they can grow and also where they can prove themselves. And this is where, as we are open, curious, and also experimental to a certain degree, we are open to it. And this is maybe a competitive advantage compared to other large companies you don't see on the sponsor list, that they say, okay, what have the others done? Where does this work? And can I buy it from a large company like myself that I could sue if it doesn't work. So that's not our strategy to a certain degree. So our strategy is find the best solution that works for the company and then scale it in a global environment.

Speaker 1 [28:51]

And so, as I understand it, you basically, you were an early adopter, you made the experience, but still made the call, we don't want to stick with my GTP anymore, rather go a better solution on the market. We pivoted.

Speaker 2 [29:02]

We pivoted to a solution on the market.

Speaker 1 [29:04]

already we will educate it with all the prior knowledge being an early adopter as well

Speaker 2 [29:09]

there as well and it might change again in a year I wouldn't know

Speaker 1 [29:13]

Basically not riding a dead horse, but just because we build it we have to keep going.

Speaker 2 [29:17]

Oh, yeah, it's a great move.

Speaker 1 [29:17]

that's a great move so how do you communicate something like this because i think many people like have spent a lot of passion into it building it it's like a baby you know and how can you say oh no we are please send it away

Speaker 2 [29:34]

Well, you really have to believe in the upside, and you have to focus on the upside. There will always be, let's say, reflexes, and there will be intuition that goes against it. But if the case is strong, and in that, essentially, what I told the team at the time, okay, we have two developers, maybe full-time on the topic, and a few others very, very part-time. They have four developers, and they're a startup. They're working day and night. They don't have family. So code quality or product quality is good. So I say, they will overtake us in three to six months. So why not join forces now? And again, it is a case by case. And ultimately, you always have to tell the people what's in it for them. And if there's not enough in it for them, then maybe, yeah, it wasn't the best decision to take.

Speaker 1 [30:33]

Interesting. So basically what it's also like just to make the right move and then sometimes sharing and joining forces is better? How can this help in Germany joining forces in open source developing software together even if it's a competitor?

Speaker 2 [30:47]

Competitors are always a bit tricky, to be honest, but there's a lot of companies in Germany that are not direct competitors. For example, Aon, with whom we are exchanging pretty extensively. So it doesn't always have to mean that you are, let's say, in a competition space. And it's not only open source, it's open source and also giving startups a chance. So, essentially, startups need large companies, and I believe large companies would benefit a lot from the startup environment, and also small and medium-sized companies would benefit a lot from open source and working, collaborating with startups. And I believe there needs to be more openness in terms of, okay, this is not, I have to buy it from the one vendor that essentially brings the standard solution to it. That is too easy, and I also believe a bit outdated in the 21st century.

Speaker 1 [31:45]

Sure, I agree. It's like, so open question for the group, because we're talking about speed, innovation, you basically just explained your strategy for speed, for innovation, the calls to make. So do we actually have enough skilled people around in Germany for really driving, building software, or do we rather the better strategy is buy? So is it make, do we have enough resources and smarts to make? or do we need to worry? What's your take on that? Just like, you can think a little bit about it.

Speaker 3 [32:25]

I could start with a thought, but it's maybe not as technical.

Speaker 1 [32:30]

Probably not also technical, maybe other skills like collaboration, communication, everything that makes the project going, everything that brings us into application.

Speaker 3 [32:42]

because you also said like Germany and what it means like for or maybe let's make it Europe yeah I think it's a question of mindset also when we look into the US if you look into China I think they have that mentality of doing it and not asking a hundred thousand questions first right

Speaker 4 [33:08]

Do you think that's good?

Speaker 3 [33:11]

Yeah.

Speaker 1 [33:12]

Oh, now it gets interesting.

Speaker 3 [33:14]

What I'm saying is, and I can say this from my own experience, it needs to be balanced at the end of the day, and my personal opinion is that in Europe there is too much weight on regulating things. Maybe that's especially true for me because I'm in the financial sector, maybe there's even more than in other industries but even get tiny things rolling like simply doing things finding out the value right is sometimes very very difficult because you have to take a lot of bureaucratic hurdles first and then when we talk about regulation it's a good thing right also i can't think we can all agree that from a european perspective the US and China don't do enough, but then the question is how do you apply regulation, right? And that is really very, very difficult in many cases when the regulator itself cannot really tell you what they expect. So you're kind of left alone on high sea and at some point in the future somebody may or may not come and tell you, oh all what you did is wrong right and that leads to the effect that you're doing way more than you should do that you're producing way more paper than you should do and all that is not creating value right and I think that is a problem that we have to solve

Speaker 4 [34:56]

No, and I think overall, I mean, we do have the general skills and the general industry because, I mean, AI is kind of an interesting field because it's becoming more and more important to have a university degree to get a job, even though the things you actually learn are much less relevant. You know, you don't need a PhD in machine learning to do something productive in the field. And I think we do have a lot of, you know, a lot of interesting businesses, business use cases, a lot of skilled people and a lot of problems. And I think, you know, we should focus on that instead of trying to, I don't know, chase after some idea we have for maybe looking at the U.S. And do we need our own open AI? Do we need, you know, do we need all these like...

Speaker 1 [35:47]

I think one of the key messages is here okay we just need like stability and clarity that's the thing because of course like imagine somebody to do code review who cannot program but sometimes it could happen with a regulator just like over exaggerating a bit and nobody wants to take real responsibility and we I don't only see this in finance I also saw it recently we have my husband in the chemical industry sometimes people yeah it works

Speaker 2 [36:12]

It works, all of the paper.

Speaker 1 [36:13]

the paperwork isn't fit but nobody wants to sign it off and take responsibility so yeah anyway

Speaker 4 [36:18]

We need to understand the technology in order to regulate it.

Speaker 2 [36:22]

Well, if I may add quickly to that one, because I would second everything that was said. Maybe a little add. I believe on the talent side, Germany and Europe is solid, definitely. On the industry side, we're also very solid. So we are still one of the highest GDPs in the world. We have a lot of, let's say, classic industry. So there's a lot of interesting application fields for AI that are just lying around to a certain degree. So we have talent and we have the application fields. but somehow still we're not leading in those things, and scale is missing. Why is that? I believe something of a catalyst, maybe there's a disconnect or something. People, when they really want to scale, and I also hear this from startups, they tell them, you need to go to the U.S. because there you will grow and you will find customers. And I believe what is missing is a bit of openness in terms of experimentation and in terms of giving the young talent a chance to get to a scale. and we can learn from the US we can learn from China but we have to find our own way to go with those things so not just taking pattern A or pattern B and try to be a me too but essentially to say what is our way what are our strengths and how can we let's say learn the mistakes the others made and avoid them

Speaker 4 [37:38]

them. Yeah, and actually maybe to add to that, like from sort of a founder's perspective, like people are often surprised when I say that what always inspired us as a company and also as a German company was the German Mittelstands or the German middle-class companies. That's something that always appealed to us much, much more than the hyper-growth start-up vibe. And I do think that is something that is quite...

Speaker 1 [38:00]

And the German middle stand were your early adopters?

Speaker 4 [38:04]

Um, no.

Speaker 1 [38:05]

No, who were you all the way down through?

Speaker 4 [38:06]

I think it's definitely, so companies that were able to move fast, so even companies in the Nordics, they have, you know, sort of a slightly different mentality. So we saw those first before we saw the German companies. Then of course, you know, startups that can just quickly make decisions, but also surprisingly a lot of...

Speaker 1 [38:27]

Not like US?

Speaker 4 [38:29]

Yes, definitely. I mean, I would still at least, you know, especially at the beginning, it's like 60% of our customers. And also because there's just more tech companies and more startups.

Speaker 1 [38:40]

So a little bit more, they'd rather take a risk, probably.

Speaker 4 [38:45]

But my takeaways no, no

Speaker 1 [38:46]

No, no, no. The companies probably take rather a risk in the U.S.

Speaker 4 [38:49]

Yeah, and you have startups that are well-funded with just a lot of money that they need to spend. But we did also see, you know, German Mittelstand eventually coming around. I think it's more of a personal, it's a personality thing as well. You have, like, some companies that seem super bureaucratic, but, like, they can move fast. They have a competent team. They're just, like, building things, getting shit done. And then on the other hand, you see these, you see U.S. companies that operate more like startups and pride themselves on, like, how they've disrupted whatever. And, yeah, they were unable to buy a license from us because, and, you know, spent weeks in legal and eventually failed for, you know, something that cost a few thousand euros and they were unable to purchase it because, well, we refused to just change the jurisdiction of our contract for a purchase of $2,000.

Speaker 1 [39:36]

Yeah, all kind of like governor.

Speaker 4 [39:37]

So, yeah, so you want to figure out, well,

Speaker 1 [39:39]

Who does not like covering an illegal flag?

Speaker 4 [39:40]

And companies, you kind of ship your org chart, so it really depends.

Speaker 2 [39:47]

So it's all how

Speaker 1 [39:49]

Some more open questions, like quick answers. How do you balance excitement of adopting open source and new stuff?

Speaker 2 [39:56]

Uh, with...

Speaker 1 [39:58]

cautious and risk culture so is it like sandbox model we don't cry try i mean you probably cannot say we just try and wait until they regulate so what's what's the approach to keep up with most to to try new things um to take probably a risk but not being the risk for the company or like breaking laws or whatever like like like how how how how navigate i think i think a lot of language models are a really good example and the data they train on there's always like a gray area what's the strategy to navigate yeah

Speaker 2 [40:35]

Yeah, more naturally to a certain degree in a global company you will always find someone somewhere who is interested and curious to try things out and sometimes you have to see if let's say the local regulations are let's say more flexible or more tight and then decide if that's a good idea or a bad idea but in essence it's piloting in a sandbox.

Speaker 1 [41:03]

More deals

Speaker 3 [41:05]

For example, what we do is we have our APIs, for example, to Azure, then OpenAI, and then that is a safe and secure path for us to utilize LLMs because it's contractually safe and all that, as it has to be. and on that basis we can build use cases right um and for example um i have i found this quite quite interesting because it seems to be a general problem we have a software with which we answer to rfp so requests for proposals right and they offer that you use your own azure subscription which is great because then we can use that and uh in our in our safe haven so to say when it comes to utilizing other lamps right so that's actually a quite intriguing thought even though it all goes then back to microsoft and open ai but maybe that changes in the future

Speaker 1 [42:06]

How do you navigate? Of course you're building an open source.

Speaker 2 [42:09]

Yeah. Okay.

Speaker 1 [42:10]

Okay, an open AI is coming with different approaches like that, do you think, okay, we just do a thing, we stability, we integrate, how do you keep up? What's like the inner workings of an open source tool like Spacey, like also like navigating innovation from outside, what's happening, not being the next legacy. Yeah.

Speaker 4 [42:32]

Yes, I think our work has always been taking what works in research and also what has proven to be effective and make that available for companies to use in a practical sense. Part of that is, of course, you want to follow what's going on, but you also want to keep a bit of a distance. You don't want to be just implementing any latest paper that comes out and get really sidetracked. You want to see what sticks. In research, that's just time. You see how do things develop and you need to be able to follow that kind of thread and see, okay, this technique has actually proven to work. For example, something like transfer learning finally works, it's effective, that is actually something. It's very nice to see all these visions that we've had since we started the company come to fruition. We always thought like, ah, someday we will have models good enough that you can bootstrap Your AI pipelines with so you only need very few annotated examples. That's something that's kind of the foundation of our vision And now it's like we're finally there You know, it was kind of kind of playing the long game there, so Yeah, definitely. I think knowing when to take a step back And also at every step reasoning about what makes sense in keeping a close connection to our actual user base because people are not only looking to us to implement things but also looking to us to for best practices and advice to see what works how do I approach my problems and again the hard part is not which algorithm or model to use the hard part is I have a business problem how can I break it down into things that are easy to solve with machine learning and that is people people don't like hearing this because it is something technology won't magically solve this like you have to you always have to think and reason about what

Speaker 1 [44:23]

What the hell you doing how to bring it in the application and if?

Speaker 4 [44:26]

And there's no playbook and you know that makes people uncomfortable, but at least that is kind of the reality of so

Speaker 1 [44:31]

So how do you keep up? I mean, you're in the middle of NLP everywhere, so how do you keep up with the latest developments that are probably not relevant, not natural language processing?

Speaker 4 [44:42]

Oh, like other fields? Yeah, other.

Speaker 1 [44:44]

Yeah, there are technologies that could also benefit spaCy or not, which are not like on the natural path, so there's always like innovation coming from different domain sites. Do you care? What's your approach to keep up with it?

Speaker 4 [45:00]

Definitely, you know, you do need to keep a focus, otherwise you kind of, you know, you have information overload, but definitely, you know, following what's going on, also keeping an open mind, and, you know, staying creative, also from, like, you know, the application and product perspective, and I feel like that's actually always something I've enjoyed, like, recently I was at a conference where I gave a talk, and I could meet up, and there was a talk that was actually, you know, kind of agents, and quite not very relevant to what we're doing, but, like, there was an idea in it, I was like, oh, okay, this, we could connect to something we've always wanted to do, and there is this new product in there, and so, yeah, that's kind of what's currently on my mind, and so that's, I think, why events like this, for example, are really valuable, and I've already, even now, I've met so many users of our tools, talked about their use cases.

Speaker 1 [45:52]

But this is like spacey, which is like, I would say, like a smaller space. Now let's, like, you have like many applications in a larger organization. How do you keep up? Do you just like, hey, team, come pitch it to me. I'll decide what's your strategy. How do you keep up with everything? Like just to see what's interesting. Where should we, what should we try? What's the approach?

Speaker 2 [46:14]

So I read, I listen, I talk to people, and sometimes I have stupid ideas. I go over to my team and they say, forget it, we're not doing this. Sometimes they say, okay, we'll have a look, and then they don't do it. And then I ask again, and then they do it. And then sometimes it makes its way into our stack, and sometimes it doesn't. Ultimately, it's about experimentation and about the inception point. So we do have a dedicated team. called the quantum in ai lab and their sole purpose and their sole task is essentially to look ahead and tell us when the inception point for an idea for a technology for a piece a component has come and that we can then have a closer look collectively experiment with it and then build it in or not

Speaker 1 [47:00]

Do you also check the other side, check like outside sources, what should my team look into, like keep just yourself informed?

Speaker 2 [47:08]

Both directions. Ultimately, I also have to be, let's say, stay a separate entity, and also stay a curious mind, even as I'm getting older, and therefore I also have my own streams of knowledge coming in.

Speaker 1 [47:23]

Oh, okay. Actually, I heard RSS feeds.

Speaker 2 [47:26]

I'm still using RSS feeds, I'm that old, yes.

Speaker 4 [47:30]

Yes, me too. I'm still I have not forgiven Google for killing Google reader. Yes

Speaker 2 [47:34]

Yes.

Speaker 1 [47:35]

So how was your strategy keeping up with all this, how to decide, like, there's 15 papers, 70 new approaches every day.

Speaker 3 [47:48]

My personal challenge really is not to get too excited about everything, right?

Speaker 2 [47:53]

Yes.

Speaker 3 [47:54]

And you need to maintain some consistency in what you do, so it's really sometimes there is more value in getting more of what is already there than jumping ship all the time, right? Of course, we've only recently decided to go with a very well-known cloud database provider that also incorporates analytics and dashboarding and everything. So I think that's a fantastic learning field for us and there will be a lot of value for us to get out of if we do it correctly and then it's also a lot of just hard work to get there and not get distracted, right? That's my challenge.

Speaker 2 [48:35]

Thank you so much.

Speaker 1 [48:36]

Oh, yeah. Thanks very much. Let's do a fun closing. Quick questions, fast answers. Agents, now or never.

Speaker 4 [48:47]

Depends on how do you define, I mean, stupid question, I mean, no, I mean, sorry. There are no stupid questions, but this is a...

Speaker 1 [48:57]

It's closed. It's closed. Questions were like, yeah, let's spark, like, nah, whatever.

Speaker 4 [49:03]

Oh

Speaker 3 [49:07]

For the right use case now

Speaker 1 [49:08]

Oh, okay, diplomatic answer, yes.

Speaker 2 [49:10]

Sure, yes.

Speaker 1 [49:11]

the best choice you ever made the best choice

Speaker 3 [49:15]

The best choice I've ever made.

Speaker 1 [49:15]

the best choice That's the best pose you ever made.

Speaker 3 [49:17]

With regards to Python, I think this really goes back to 2020 when we said for the whole company, we are a Python company.

Speaker 1 [49:28]

Thank you. We like to hear this at Python conferences.

Speaker 4 [49:33]

Sorry, I'm still thinking about agents in a minute.

Speaker 1 [49:37]

So probably just all go up to your feminist AI room and maybe hack some agents because we have the hardware and everything there and we have the right people there.

Speaker 4 [49:47]

I'd say, I don't know, I have a problem with the way, you know, the messaging around it. There's, of course, good technology there.

Speaker 1 [49:53]

there. We're talking application. Yeah. Yeah. But best decision.

Speaker 4 [49:54]

What are we talking about? Oh, I think in general, going into the developer tools space, I do think enabling developers to do stuff is very satisfying.

Speaker 1 [50:09]

Well, it's your best decision ever made.

Speaker 2 [50:11]

Best decision ever made, my wife, my kids. You didn't specify better.

Speaker 1 [50:16]

No, it's a valid answer, everything. So beyond hype, agents, so forget about agents for a minute. No, no, forget about agents, we never talk. We don't know what agents are. They're like, yeah, beyond hype, which technology should we really look into? Like for long, mid and long term, beyond all the hypes.

Speaker 2 [50:40]

worse

Speaker 1 [50:41]

What's the thing, what's the signal and the noise currently?

Speaker 2 [50:41]

worse

Speaker 3 [50:55]

Again, I would come from a bit of a different angle. I think now we have that global situation with Donald Trump, et cetera, trade war maybe looming. We all don't know. I think everything that's coming out of Europe, especially Germany, is going to be worth a look.

Speaker 2 [51:12]

I still believe there is a beautiful competitive advantage in merging hardware and software. So looking into, let's say, a strong software design with a custom hardware chip design. This is something that I believe Apple has played very nicely, more and more companies like Grok in the inference space, and also I forgot the name of the European companies also competing in that space. But let's say doing an integrated view is a heavy investment, it's a hefty investment, but I believe it's also a high risk, high opportunity play. I would love to see an European player doing this.

Speaker 1 [52:00]

Hotware

Speaker 2 [52:01]

Interesting. Hardware and software, the integration.

Speaker 4 [52:04]

Yeah, I think maybe in a similar vein also in general product because I do think product decisions are ultimately what's gonna Drive the innovation forward and the use cases and not just the technology because there's so many decisions you have to make around how should innovation look and You know, it's like I think an example always like to use is this the window knocking machine You know back in the day people would hire a guy to walk around knock on people's windows and we have to replace that with with technology but we didn't build window knocking machines we built alarm clocks and i think if we go one step further beyond just trying to translate human tasks one-to-one

Speaker 1 [52:40]

And we have to also start just to replicate what we know into the process. Yeah, like taking, we have technology, like we have many, many tools, and we just probably just think, hey, how can we build from these things and not just like rebuild what we have? And I think that's a great closing. And I would like to hand over to Florian for the questions.

Speaker 2 [53:05]

Okay, here comes the Slido agent. Ask your questions, dear Slido. Thank you so much, first of all, for the great discussion. Question to the panel. How do you enforce coding guidelines in your company, especially in teams with varying...

Speaker 3 [53:21]

with varying programming skills.

Speaker 1 [53:22]

Programming skill levels

Speaker 2 [53:33]

Nicholas, are you out there?

Speaker 3 [53:34]

Ha, ha, ha, ha, ha.

Speaker 2 [53:36]

You're not out there, or we don't do it? So it's an excellent question. It is actually something that's a fine line to walk. I believe with more and more Gen AI coming into the coding environments, this is something that can ease. So ultimately, I'm not a strict, let's say, believer in narrow standards and terms. There's only one way to do it. So I like to think of it as a freedom in a frame. As long as you do decent documentation and you, let's say, don't write crappy code that costs us thousands of euros in the run environment, I'm OK. So I guess it is a question of balance. And ultimately, I hope that in the future also, GenAI-empowered IDEs can support better coherence and better quality of code and better documentation.

Speaker 1 [54:37]

also easier I mean if I remember like four years ago we had to write a lot of guidelines rules and if you basically do it today you just say let's use you we let's use rough because also here we are at a completely new level because great people build great standards they're easy to follow and basically there's zero discussions anymore about coding styles on this because they were never overproductive.

Speaker 4 [55:03]

I mean remember when you had to like actually hadn't format code like before black and stuff where you're actually like 90 characters

Speaker 1 [55:10]

And now I think that's also like this is like something nobody would basically celebrate this late as a big like a big innovation Yeah, it's probably not rocket science, but it's super useful and a real speed up for large teams We save a lot of time with that

Speaker 3 [55:27]

Maybe in addition to that I mean you have a development process right that may or may not hopefully it does include code reviews but it goes further right so you will have maybe vulnerability scannings at some point in your process etc etc then reveals if someone has hardcoded the password right.

Speaker 2 [55:50]

Okay, great. Thank you for your answers. A question to you. Do you have the statistics how many AI ideas or use cases made it to production at your company?

Speaker 3 [56:03]

I don't want I don't want to make anything up actually I cannot say we are 100 people a company I would say we have at least five very good use cases that I know of maybe ten more in the pipeline that have potential I think that is that is

Speaker 2 [56:27]

Thank you. And then next question. What do you think about the effect of AI on production processes based on 3D printing?

Speaker 3 [56:34]

3D printing.

Speaker 1 [56:40]

It's a very specific question.

Speaker 2 [56:43]

Yeah.

Speaker 1 [56:46]

I don't know, you're not into 3D printing.

Speaker 4 [56:52]

It's kind of cool, like I've always wanted like a 3D printer.

Speaker 3 [56:55]

Maybe the question goes into that direction of you can use AI to work with Blender, for example, to render some 3D models and then put that in production. Maybe that is the question, but also if that were the question, I couldn't answer it.

Speaker 1 [57:12]

I would say let's skip the question if we cannot.

Speaker 2 [57:15]

So what I do know that, for example, talking about German Mittelstand, that the company Trumpf is using a steel cutting machine that is using AI to optimize layouts for steel cutting machines to reduce waste, for example.

Speaker 1 [57:15]

Sorry.

Speaker 2 [57:32]

So you don't do manually the design anymore, but AI optimizes and color codes the different parts and how to split them up afterwards. so you have less waste in it. So that's one application. Thank you so much. As the moderator, I inform you that we are out of time.

Speaker 1 [57:51]

you that we are out of time okay thank you very much thanks everyone for joining thank you for being around thanks for having us thank you thank you so much and yeah enjoy the conference

Speaker 2 [58:01]

Thanks for stopping by. Thank you. Thank you.