1. Podcasts
  2. EUVC
  3. Andreas Cleve (Corti): Why constraints can be an advantage in vertical AI
EUVC

30 Oct 2026 · 50m · 26 chapters

Episode artwork

Andreas Cleve (Corti): Why constraints can be an advantage in vertical AI

In short

Going vertical in healthcare AI by building an API/infrastructure platform focused on compliance, accuracy, and real-time clinical workflows rather than “horizontal” general-purpose models.

Guest (Andreas Cleve / Cordy)

Co-founder of Cordy (Copenhagen). Started in 2017 as a healthcare research company; ran a clinical RCT proving low-latency real-time AI for clinicians. Pivoted to commercial APIs during COVID; raised $100M; now a healthcare AI developer platform with ~2,000 builders and ~12–15% week-on-week growth. Competes on healthcare model quality within the domain.

Key claims

Healthcare is “many markets,” not one; small errors (even one word/letter) can change outcomes. FDA/compliance constraints make vertical focus an advantage. Accuracy requires “scaffolds” and deep integration with clinical language/coding systems (e.g., SNOMED/ICD) and edge workflows. Regulation is a necessary standard for patient trust.

Notable examples

radiology’s low tolerance for mistakes; ambient scribes for outpatient dictation; enterprise EHR rollouts requiring massive compliance across jurisdictions.

Written by AI. May contain mistakes.

Chapters

0:00

Market Dynamics in Healthcare AI

Explore the shifting focus in AI from larger models to specialized healthcare applications.

“But I think that sort of evades the market a little bit here pre-IPO frenzy in the US.”
0:45

Introduction to Cordy and Its Journey

Learn about Cordy's transition from a research company to a healthcare AI developer.

“I think our compliance team is bigger than our HR team.”
1:47

Founding Ethos and Market Specialization

Understand Cordy's focus on healthcare and the need for specialized AI solutions.

“To those that don't already know you, please give a quick intro to QWERTY, and then we can jump into the thinking behind going vertical.”
3:15

Building Real-Time AI in Healthcare

Discuss the challenges and opportunities in developing AI for real-time healthcare applications.

“We released the building platform last summer, 25.”
5:08

Trade-offs in Going Vertical

Examine the trade-offs and challenges faced by Cordy in focusing on a vertical market.

“that we thought somebody has to focus on this full-time because this model never got there.”
7:05

Investor Perspectives and Market Dynamics

Explore how investor expectations influence Cordy's strategy in a competitive market.

“Did you have a time where you felt like this is costing us too much in terms of, one is that we all know the importance of being able to build a narrative that allows you to raise.”
10:00

Fundraising Journey and Capital Challenges

Delve into Cordy's fundraising efforts and the capital demands of AI development.

“The challenge for us was, first off, we were both building a nice application SaaS business and we were building our infrastructure.”
11:41

Talent Acquisition in Europe vs. the U.S.

Discuss the challenges of attracting talent in Europe and the potential need to expand to the U.S.

“Most would say that it's not that it's too complex, but it's just that we don't have the depth of capital.”
13:20

Future of Work in AI Development

Consider the hybrid work model and its implications for building a successful AI organization.

“So commercial headquarters where I sit is in New York.”
14:00

Copenhagen's Tech Ecosystem Challenges

Explore the historical and current challenges in Copenhagen's tech talent landscape.

“So as you said, talent is there, but I don't think a lot of people have been on a growth journey as they've been lucky to do more in Stockholm and London.”
Show all 26 chapters
15:00

Building a Hybrid Organization

Learn about strategies for building a hybrid organization to attract talent.

“And there is a distance at least that's created.”
17:44

Navigating Talent Acquisition Issues

Discuss the difficulties in attracting tech talent to Copenhagen.

“There's obviously great AI companies, but you might not know them, right?”
18:28

Importance of Accuracy in Healthcare AI

Understand the critical need for accuracy in healthcare AI applications.

“So understanding these LLMs, what goes on inside, find it on hogging face and all the way through our latest launch in so agentic ready speech to text.”
22:44

API Infrastructure and Complexity

Explore the complexities of being an API infrastructure provider in healthcare.

“We cannot do the SaaS model side of this because it is too complex.”
24:47

Adoption Trends in Healthcare Technology

Examine trends and challenges in the adoption of healthcare technology.

“I think where we see our job a little bit differently than the horizontal providers is obviously we can hire and build specifically for clinical reality and we can get much deeper in clinical reality.”
28:01

Navigating Compliance in Healthcare

Learn about the complexities and compliance challenges in healthcare AI.

“That's where we spend a lot of our time because the compliance standards for rolling out something at that scale is massive.”
28:35

The Economics of Healthcare

Explore the financial dynamics and market opportunities within healthcare.

“Obviously, that probably doesn't work with the big ones in this space.”
29:52

Challenges with AI Adoption in Healthcare

Understand the barriers to AI implementation in healthcare systems.

“And I think in some industries where there is zero barriers to large language model providers like Claude and, or sorry, Entropic and OpenAI, we already see that play out on public markets.”
30:57

Rethinking Healthcare Regulation

Gain insights into the necessity and impact of healthcare regulations.

“But probably they won't start with the clinical reality.”
33:12

Approaches to Navigating Regulations

Discover effective strategies for startups to handle healthcare regulations.

“It's probably going to stay permanently.”
34:20

The Path to Product-Market Fit

Learn about achieving product-market fit and the current VC landscape.

“seeing them as first and foremost instrumental for giving us a safe healthcare system.”
36:39

Managing VC Relationships

Understand how to effectively manage relationships with venture capitalists.

“but you would always say that you would want to control and own the success in any space that's absolutely instrumental to you.”
39:21

Trends in VC Behavior and Strategy

Explore the evolving behavior of VCs and their impact on startups.

“It's a critical service industry if you're building in deep tech.”
42:00

The Evolution of Venture Capital in Europe

Explore the changing landscape of venture capital and the importance of GPU strategies.

“But I think you're absolutely right about the importance of optimizing for the best names and the people that really know what you're doing.”
45:07

Challenges Facing European Startups

Discuss the obstacles European startups face in scaling and going public.

“I completely agree it's not something I've seen in any way being led by a lot of European VCs.”
47:37

Deciding on U.S. Expansion for Startups

Understand the considerations for startups contemplating moving to the U.S. market.

“How do you advise fellow founders about going to the US because it's obviously one thing many have to do for the commercial side, some do it for the fundraising side.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On EUVC, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00But I think that sort of evades the market a little bit here pre-IPO frenzy in the US. Everybody's just talking about bigger models. But although the models haven't got like an order of magnitude or two bigger, they're not closer to be FDA compliant, right? Whereas today, that same hospital might have a surgery for a very, very small body part. That's specialization. That's great because that means we all get better health care if we can get access. So the thesis here is very much health care is not one market. It's many markets. They're billion-dollar markets. They're very deep. If you are spilling one letter, one word wrong, the entire situation changes.

0:33That's created a situation where 90%, kind of good, looks great in demos. What you can then do is what the best application guys are doing out there. They're building massive scaffolds around their platform to make sure that their exact core market can manipulate the AI to get to a certain space. I think our compliance team is bigger than our HR team. My advice would just be, I know it sounds daunting. It is really not that daunting the moment you get out of the advisory industry that wants to tell you it's the hardest thing in the world. It's definitely much easier than building your company.

1:06It's actually taking it upon yourself to own it as much as you can and find the path of least resistance. And it's probably the path with least amount of people involved. In 2026, every AI company gets the same advice. Build the platform. Be the horizontal layer. Maximize the TAM. The guest today went the opposite direction. Eight years ago, he co-founded Cordy in Copenhagen as a healthcare research company. Today, it's a healthcare AI developer platform. 2 ,000 builders on the platform, 12 to 15 percent week-on-week growth, 100 million raised, competing directly with OpenAI and Mistrala model quality, just in one domain, healthcare.

1:46Andreas Cleve: This show is not investment advice, and the hosts of this episode may be invested in the funds and companies featured. To those that don't already know you, please give a quick intro to QWERTY, and then we can jump into the thinking behind going vertical. Yeah, great. Thanks for having me. We are a healthcare developer platform. We orchestrate models, train models to models, and deliver agentic infrastructure to help AI become table stakes in healthcare. Quick stats on you. What have you done before? How much have you raised? I think the big important thing about Europe is that most people don't know the incredible champions we have, and you're one of them.

2:22Thank you. We started in 17 as a research company. Our 17 thesis funded by Sunstone, and our hard call was context over language models, spent three and a half years in research and a clinical RCT, proving we could have low latency. This is a different era of language models, but language models running low latency in real time to help clinicians make better decisions. You can find the research out there. Cordy became into being as a commercial company selling APIs during COVID, build a nice double-digit AI SaaS business around our own models because nobody wanted APIs or AI company from Copenhagen during COVID.

2:58And in 23, we started seeing more and more traction around our APIs. ChatGPT moment had happened, and we could get back to our ethos and our founding sort of thesis of building infrastructure and training models. So that's our thesis. We're growing, I don't know, 12, 15 % a week. We have around 2 ,000 builders building on Cordy. We released the building platform last summer, 25. What we help with is if you want to have clinical-grade AI running in production or you're building for clinical-grade type AI workloads, you're probably going to want to have help from somebody who sees thousands of cases, have compliance bespoke models that actually work for healthcare, that thinks about nothing else.

3:36And that's exactly what we do. Beautiful. Andres, tell me, I want to dive into this decision to go vertical at a time where it felt like everything was about scale and thus also you wanting to go horizontal. So dive into that. What was the first thinking around why you should go vertical? So our sort of founding ethos, obviously, from all of the way back to all of the people out there who remember the deep learning days was very much trying to find a data modality that really served us well as we started to build deep models. So we need a lot of data. We need it quite well labeled and quite well understood.

4:12And obviously we wanted to build a company around healthcare where I had built a company before and my co-founder was passionate that had the thesis that we thought would be the most impactful, which was there's not enough clinicians. There is simply not. And the quality of access to clinicians is also worsening, even in the best parts of the world. So too few people, too little access to them. And us as consumers, we just want more, more healthcare, more health, more all of it. And that's sort of the supply-demand dynamic we saw when we found it in 17 would get worse and obviously has got quite a lot worse.

4:45So the thesis was, could we build real-time AI that actually understood real patient conversations? And what we just learned quite fast was that general purpose is great. Like healthcare will have a great time using general purpose models of 2026 for healthcare marketing and beyond, and a lot of great use cases there. But healthcare is so deep if you look at the size of the market and the need for really governable, compliant AI that we thought somebody has to focus on this full-time because this model never got there. If we start with what it gave you, then we can go afterwards into what you had to give up.

5:16But what did this decision give you, so to say? What did it enable you to do that no one else could? Yeah, so obviously, first things first, what we deliver is an API, it's a platform, console studio, like you have on OpenAI. We don't see ourselves as competing. Our job is not to build the best horizontal models, to make sure you actually can get something as good as you would want, so you can get it live, because healthcare mostly don't have AI live. So what it gave us was a chance to focus. So I think the dominating narrative of AI today is that train time compute, so pre-training and more data will solve all the problems of the world.

5:53And if you really study what has transpired so far, at least in the current paradigm, is that although we're plowing many, many billions into these models, they actually obviously become a lot better. A lot of the latest ooms, as Leo Paul-Echenbrander calls them, So order of magnitude improvements, they actually come from not from pre-training, but from a mixture of other things that compound together. So I think reasoning models, the 03-01 moment a couple of years back was a real moment. I think a lot of people didn't actually see that it wasn't about a new model. It was the same model as before, but it worked now as a reasoning model, multi-model and multi-expert interactions.

6:29And this mixture of experts and much, much more has been compounding layers on top of great pre-trained models. But in reality, as we see today with harness engineering, controlling models for specific paradigms for great outcomes is a lot more about understanding the realm you're in and understanding the models and understanding how to navigate them versus just plowing more GPUs into pre-training. But I think that sort of evades the market a little bit here pre-IPO frenzy in the US. Everybody's just talking about bigger models. But although the models haven't got like an order of magnitude or two bigger, they're not closer to be FDA compliant, right?

7:02Yeah. So that's what it gave you. Then what did it cost you? Did you have a time where you felt like this is costing us too much in terms of, one is that we all know the importance of being able to build a narrative that allows you to raise. You're raising in a market where everyone's talking about go horizontal, you've got to go bigger, and you're doing the opposite. Did you feel at times we probably should adapt our model to allow us to raise more so that we could keep pace? Again, so we are an AI model provider. We compete with other models coming from very, very big horizontal labs. So first things first, when you only focus on one market, you only have that total addressable market, right?

7:41And obviously, we believe healthcare spends more than a trillion a year in the US alone on admin. That's sort of the easy market. More importantly, it's just we believe there is so much more to be spent on just us getting more healthcare. And I think reasoning models and beyond will play a big role there. So we believe our market was big enough. But obviously, if you don't invest in healthcare, the investment thesis falls short already there, right? Secondly, we believe that it's not going to be a one model solves it all. We believe that healthcare is actually not one market at all. We believe it's many, many markets.

8:09So radiology is so, so, so different from dentistry or psychology. And although we'd like from a technology perspective to standardize and treat them all the same, you should not. And we can talk later about this being, I believe, the hottest professional user group in the world to deliver software to. But it's not because they, as one, are hard. It's because they and their specialty are specializing more. So if you go back 80 years to a rural hospital in mid-America, they might not have a lot of surgeons. And those surgeons will do a lot of different types of surgeries. Whereas today, that same hospital might have a surgery for a very, very small body part.

8:43That's specialization. That's great because that means we all get better health care if we can get access. So the thesis here is very much health care. Health care is not one market. It's many markets. They're billion-dollar markets. They're very deep. So yes, you're giving away TAM. and next up you have to get very deep in the tech stack if we just came here and said hey the only thing you can build on cordy is one of these ai scribes we wouldn't last long because you need to actually believe you can have your entire roadmap running on cordy and you have ambitions right so our our sort of attack vectors the surface you have to cover doesn't feel very narrow when you sit inside healthcare because we have to cover all the basis of you okay you want to build agents here's an agent framework that's better than than lane chain okay you want to build text generation.

9:25Here's text generation that's much, much more trustworthy than what you get on OpenAI. Ah, okay, you want to build speech-to-text. Here's something that can beat 11 laps. So our sort of spectrum here is quite wide. So we're giving up TAM for a very deep battle, controlling all the way down to software infrastructure or to GPUs, right? So a lot of risk assumed and a lot of work to be done, and obviously a TAM-constrained situation. Were there any places where you would say that going vertical felt like we can't do it, it's cutting off too much? Or how did you approach that in terms of the trade-offs and how did you talk to both your investors but also your founding team about it?

10:07The challenge for us was, first off, we were both building a nice application SaaS business and we were building our infrastructure. So our own SaaS business was running on our own infrastructure, our own models, and so were some of our customers. And the first things first was obviously we had to tell our investors and the parts of our employees that the customers that was running on our SaaS products, they were going to run on hopefully partners or builders running on Cordy in the future. We were not going to build, keep building on that SaaS platform. And that obviously is a new dynamic. If you are a VC with the thesis of investing into software as a service, we're saying we're going to build the model layer, the infrastructure layer, and it's a different thesis, right?

10:45Secondly, there is a completely different need on resourcing. So obviously we have always trained models back to our founding, but God knows there is a big need for control and we have a quite large portfolio of offerings today. So a completely different approach to talent acquisition, where to build talent, how to track them, and then GPUs. Do we have infrastructure enough? Do we have the infrastructure to partner with from the hyperscalers? And what's the sort of GPU per head we have access to so we know that each brilliant individual can be meaningfully creative, right? So there was a lot of sort of navigation of who we are and what that means.

11:21And I think for many in Europe, there was sort of, I think there have been for the last, what, I don't know, 8, 10, 20 years maybe, this feeling that, oh, this is tough, right? Probably not for Europeans, right? Yeah, or at least not this is tough, but this is too capital demanding. I think most would say we do have the talent. Most would say that it's not that it's too complex, but it's just that we don't have the depth of capital. Maybe you can talk a bit of that capital fundraising journey. You've raised massive amounts, and you're obviously on a path that will require even more money raised, and you're doing it out of Europe.

11:57Can you comment a bit on that? Yeah, of course. So we've raised$100 million, which I find very slim. I think we're at the most, both per head and relative to revenue, I think we're the most lean AI lab, maybe among some Chinese ones in the world. And luckily, we have great partnerships with NVIDIA and Microsoft, so it wouldn't be doable at all. So our core innovation is a lot around the test time compute thesis, the reasoning, the understanding of the market, the multi-agent space and building and tuning and training there. And obviously around a large swath of data and insights and a great data pipeline, a synthetic pipeline that we've built ourselves that allow us to innovate.

12:36as we've sort of proven and that was the thesis of the year if we can become state of the art so if we can become soda on the core infrastructure of medical reasoning and control the medical intelligence layer then we're gonna gonna invest a lot into that thesis which is something we're gonna announce much more about this autumn so so yeah it's a capital intensive journey right we are a lab and and we train models and and obviously health care is a deep as a market as i said so So, yeah, it's a completely different journey. And obviously, we're somewhere where we're going to look very internationally to accrue enough capital to make that happen.

13:12Are you building as much in Europe as you would want to? Or are you starting to see that you need to move to the U.S. more and more? So commercial headquarters where I sit is in New York. We have our commercial team as well and back office team in London. And then we have R &D in Copenhagen. And we've been worried as hell around whether or not we can keep doing that, not because of the talent. Talent is crazy good. But I think we're, I don't know, 20 nationalities in Copenhagen, sub 100 people. And we're very bound still to Copenhagen continuing to be an amazing place to live so we can attract people from all over the world to come and live.

13:45And then obviously the legal paradigms softening to recruitment of our style. So that's at least something I worry about generally for Europe is if we're heading in a direction where talent acquisition increasingly will become easier because we definitely need that. We've actually chosen to double down on Copenhagen and Europe. So not only do we have one of the best tech universities, if you go back through the history, PHP or Bluetooth contributions and so on, there's so many great scalps and milestones coming out of the tech ecosystem and before that, the physics ecosystem in Copenhagen around the Sport Institute.

14:17So as you said, talent is there, but I don't think a lot of people have been on a growth journey as they've been lucky to do more in Stockholm and London. So there is not a lot of sort of, especially commercial talent and talent who's used to scaling things at a large scale. So that we obviously have to go find elsewhere.

14:34First question, are you trying to build a hybrid organization where you're going to be able to attract talent from across Europe that then maybe once a week or something like that have to come into Copenhagen? Or are you trying to build an organization where everyone has to be in Copenhagen and that's for R &D as an example? We'd like everybody to be in Copenhagen more, especially if they're building. But I also think it's just a pure fact that if you are 100 people or you're sitting in two floors and the moment you're on two floors, you're remote. You don't meet everybody every day. And there is a distance at least that's created.

15:08So I don't think we're sort of absolutist. You have to have eye contact a couple of times a day. But I think for creative work, especially on the emergence of capabilities on these models, it's really, really meaningful to have everybody in a room every day. So certain teams were pretty religious and other teams were much more relaxed about where they were from. How do you balance this, Andreas? How do you think about building up the capacity in Copenhagen and the environment that will allow you to bring the best from Paris, to bring the best from the many of the Eastern European countries are obviously leading when it comes to developing talent in this space?

15:47I think first and foremost I think it's it's a lot about obviously do they feel our our agenda is ambitious the target's ambitious the goal is ambitious that we want to compete with the best generally people we like to attract are people who want to compete who want to see how far they can take it but want to stay and live and build from Europe or have a clear thesis of wanting to help build something in healthcare so most employees at Cordy are have a healthcare experience in their backpack that wasn't great and have an ambition of contributing. So obviously we're hammering a lot on like that's a place to come where you get GPU per head, you get access to fantastic colleagues who can work everywhere, but they're here because they believe in the mission.

16:25And then it's trying to make sure that we double down on making an ecosystem around not only Cordy, but the AI ecosystem in Copenhagen and the office to make sure you land not just with us, but with everyone else who's thinking about building something big with AI. Are you finding it generally difficult to bring people from the rest of Europe to Copenhagen? Or are you finding that, no, you know what, they are driven by the mission enough. They're driven by what we are building. So we're not yet there where it's a problem. I believe Copenhagen specifically has a massive talent acquisition problem.

16:58So when I'm in the US, everybody knows Copenhagen. Everyone I want to go visit, and it's known as a fantastic place, could do a sabbatical in Copenhagen. But I don't think we have an environment currently where the tech industry is booming. I don't believe our local champions stayed or built deeply long enough from like Unity to SendDisk, fantastic founders, fantastic companies. I could go on. There's quite a few, but they build elsewhere. And I think that's one of the most fascinating things about Stockholm is like how long and how big Klarna and Spotify and so on has built and stayed in Stockholm.

17:26So I think the ecosystem is really lacking. And that's obviously an issue if I'm a builder and I know I'm moving to Klarna and Stockholm. I know there's four or five other companies I'd love to work for around the corner. So I'm okay moving from wherever to Stockholm because I know I'm landing in an ecosystem. Whereas in Copenhagen, if you want to work for an AI lab, you can come to us. But where would you go next? There's obviously great AI companies, but you might not know them, right? So I think Copenhagen has that problem. I don't think we have still to accept the fact that technology and increasing demand will, I think, evaporate through our industries.

17:56So Copenhagen is still very anchored in life science and pharma, to some extent logistics. and I don't think yet sort of the excitement around AI has taken a foothold. People are still very wary about the impact of AI. They're, I think, conflating it quite a bit with social media and the worries that's created. So I think Copenhagen's environment right now is tough in general for building a big tech company. How do you navigate it as a founder, recruiting? I think we focus on what we can control. We've been lucky to prove that our technology is top-notch all the way from our big launch last year on mechanistic interoperability, So understanding these LLMs, what goes on inside, find it on hogging face and all the way through our latest launch in so agentic ready speech to text.

18:39So I think across the stack, we just really try to make sure that each team are state of the art and they compete with the best globally. And then we focus a ton on just really, really hammering on every problem that is real for health care. So our builders feel that we're abstracting away some of the difficulties of building. And that attracts people and we control that. Then obviously we're going to raise more, get more GPUs per head and keep on trying to at least guide the market forward. And I think that at least resonates with a ton of builders out there. Okay, now I want to go back a bit to understanding how you're building for accuracy in a space that is notoriously difficult.

19:15Maybe you can talk first about the importance of accuracy in the healthcare market. Obviously, there's very little room for mistake in the surgeon's room. But I imagine that it's the same thing across the sector. Yeah, and I think to a lot of people it would come as a shock how difficult different healthcare is to other professions that would be deemed liable, highly detail-oriented industries. So we were last week together with one of the fastest growing companies in the world there in the legal space. And we talked about how the legal space is very different from healthcare. So if you are creating a synthesis or a summary of a conversation, the lawyer might think, it's pretty good.

19:56Like, I don't like your commas, but I'm fine. Whereas a doctor would be like, oh, that one comma, that's wrong. Can't use it. And why is that that important? It's because everything in healthcare is based on a language that people actually don't know exists. Healthcare has its complete own language. So there is more clinical terms in one of the many coding standards than there are words used in English. So if you take one of the big coding standards called SNOMED or ICD, these are types of abbreviations and coding systems that allow other people in healthcare to understand what happens with the patient.

20:25If you are spilling one letter, one word wrong, the entire situation changes. That's created a situation where like 90%, like kind of good, looks great in demos. What you can then do is what the best application guys are doing out there. They're building massive scaffolds around their platform to make sure that their exact core market can manipulate the AI to get to a certain space. It's obviously very expensive. It's very hard on margins, and it's very tough to control this massive, a little bit shaky scaffold, right? In our case, we're an API infrastructure provider. So when you come to us, you have to believe that you can get to state of the art yourselves.

21:05And many of our customers are all the way down to real clinicians building for themselves. and obviously they haven't spent five years at me after their phd from mmit on ai so they're like not comfortable with many of these contexts right so not only do we have to prove which is why we benchmark a lot publicly that we you can come to us for for state of the art but we also increasingly are learning how much we need to co-build with you to get stuff right get stuff live and leverage all the learnings we're getting from thousands of builds because one company building for radiology with us which is one of the toughest verticals as far as i see it on accuracy These people will have no tolerance.

21:39And they all do their abbreviations, their terms, the way they document a little bit differently. They won't just take the basic scribe 101 standard and say, that's great, I'll refund that. They'll need something very, very concrete because they're sitting every day for hours talking to a screen with images and needs to be on point every time. So to build with them, we're leaning much more into it. We're allowing them much more bespoke opportunities. We're building a ton of software around the models, which obviously general purpose model providers won't. They're specifically aligned with like, how do you tell your EHR platform in plain voice, although they're very different, that it should do certain things while you're talking to a screen full of radiology images.

22:17It's a very concrete engineering problem on top of these very jagged LLMs. So there is just a lot to do to get you the accuracy that's way beyond training, way beyond data. It's about making sure you can embed it into a reality that is completely different from one radiology to the next or one radiology department to the next department or radiology to pathology or pathology to psychiatry, right? Am I right in saying that you just described perfectly why you ended up going for the pure API foundational model layer and said we cannot do the UI side? We cannot do the SaaS model side of this because it is too complex.

22:58Each case will be too complex to solve at the same time as we're trying to be state-of-the-art on the foundational layer. Absolutely. And to be absolutely honest, we look at some of the people building on Cordy and some of the coolest companies and startups out there who are also building the app. They're not building the infrastructure. They might be on us. They might be on OpenAI or Gemini or Mistral or whatever. And sometimes you really envy them. oh because the last mile of like how do you highlight oh the microphone that app developer have used to record something probably shouldn't have put it behind the computer right you should probably should have put it in front of the patient but we don't control that right we don't control how it's told the story how it's orchestrated and obviously sometimes you want to you want to control more every time we've interfaced one of these problems covering i don't know 50 specialties in health care today that all expect us to be the best in the world at anything from psychiatry to up statics, we have to learn how to make everything a product problem because we don't get to be there in the room.

23:57We don't get to be there in the last mile. So now we have an API endpoint that allows you to actually check if the microphone you bought for your customers, if you are an AI scribe in healthcare, if that microphone quality is high enough. Obviously, we're not building microphones, but we learned that a lot of the build wasn't just about becoming state-of-the-art. It was becoming the easiest place to build and the easiest place to get things approved. If you don't solve those two too, you can have all the token mixing going on in the world, but you won't get there. Am I right in describing your model a bit in terms of go-to-market and creating customer success similar to what Anthropic did as an example with forward deployment engineers or whatever they call it?

24:39That's the same model you have with your partners? Yeah, we're all learning from Palantir, right? Palantir is the goat of getting complex data products live. I think where we see our job a little bit differently than the horizontal providers is obviously we can hire and build specifically for clinical reality and we can get much deeper in clinical reality. And we actually want everybody co-building with our customers inside Cordy, whereas I think some of the AR labs have such a scale that they're building these external vehicles or companies with a lot of funding, a lot of model access to go build.

25:12I think what we really cherish is the intimacy we can have with our customers and the proximity to their problems. So we have them all inside, but obviously we're all, I think, building on sort of what did Palantir learn. It's that the vast majority of technology success in deep data-driven products is happening at the edge. And given that we're not allowed to build a product at the edge, I think the next best thing is sitting at the same screen sometimes when the customers go live. I'm going to ask you a question now that's probably going to be difficult. Can you talk a bit about adoption across healthcare in general?

25:42Where do you see movements happening fastest, that type of thing? I get a question asked a lot from VCs, which is sort of what's happening? Obviously, can you tell me a bit about what you see at the data layer? And we think about a hierarchy. There is the big cornerstone of admin. That's where most of the action is happening. Most of the admin action is happening mapped directly to the workflows that's happening the most. The workflow we see happening the most are what's always happened in healthcare. Interviews with patients and dictation. That's where we see a lot of action. So a new modality in interviews is these AI ambient scribes who are almost always outpatients.

26:22So you go to the same doctor sitting at the same desk. That desk is increasingly having a microphone. But dictation still is massive. So doctors with small machines or talking to a computer or a telephone, dictating. That's massive. But this whole bundle of like, how do I capture? Massive, massive startup hotbed. Increasingly invaded by the big EHRs, the big classic healthcare companies. On top of that, we see the first sort of realm of these augmentation products. So it's much more about using an LM to write a note or write an email or summarize something or make a patient transfer note or send a referral.

26:59So they're happening on top of these, like layer one admin workflows for capture, but they're increasingly being adopted, especially in the US, slowly in Europe, but the curve is crazy. And then finally, we're seeing the first adoption of agents. Agents being, instead of sending a referral note, shouldn't we just refer them? Like, what would it take to refer them? Or do I have authority prior to be able to give the patient what I want to give them? Can an agent go check and do that? I think the vast majority is happening at the bottom still, but the vast majority of value is going to happen at the top.

27:31Yeah. And from a partner level, is it primarily startups? Is it primarily scale-ups? Is it primarily the established players that are now moving really quickly? So I think in our case, I think one of the benefits of being quite a lot of people on the ground is that we can work with some really big players. So that's where we spend most of our time is with the big labs, companies, EHRs, RISC, PECs, whatever they're called, all the abbreviations of healthcare. So imagine a big system who has hundreds of thousands of specialty doctors or nurses or whatever. That's where we spend a lot of our time because the compliance standards for rolling out something at that scale is massive.

28:06And they probably have multiple countries, multiple jurisdictions. That's where we spend most of our time. Obviously, we have a PLG motion. Everyone can go sign up right now. You get$50 free. And we have a ton of small healthcare startups who are running free. We also have an incubator for them. We spend less time there. And many of them are much more tech-enabled and obviously have a much simpler life because they're smaller. they don't have to ask as much for permission. But we work on both ends. Time spent mostly on the enterprise guys, but I think the volume is happening on the startup level.

28:33What have you learned from navigating these larger enterprise partners, both in terms of how you get into them? One is PLT. Obviously, that probably doesn't work with the big ones in this space. Another model of go-to-market is, of course, the forward deployment that we spoke about. I'm also curious to understand, are you finding that this is the land of milk and honey when it comes to readiness to pay? Or are you seeing no other being quite conscious about cost as well? So I think healthcare in volume is one of the biggest markets in the world, but in margins it's low. So there is no place in the healthcare value chain that has flush amount of cash beyond farm and life science different beast so i think generally nobody would tell you that you're going to have an easy time selling to anybody in this value chain but in volume they obviously many of them have a lot to spend right a hospital is like a large volume revenue organization right so if you do it right there's a lot of volume and you can have a great company and build great margins if you build something really meaningful but meaningful is very different and i think that's why many of the fastest growing companies are coming in as as challengers saying that we're going to replace place what's already there because we can control that ecosystem ourselves.

29:46What we've learned a lot with these big players is a lot about, I think, what everybody's trying to figure out is like, what does the future look like post-AI? And I think in some industries where there is zero barriers to large language model providers like Claude and, or sorry, Entropic and OpenAI, we already see that play out on public markets. I think in healthcare, it's still like verdict is out, like how, what is the job? How many builders? What will they own? Is EHR a thing? Is it not a thing? is it headless or is it like a software piece with like clicks and and cursors right and i think that's still many bets on it i think increasingly what we're seeing is that the especially in europe the contracts are long the changing costs are crazy so get a hospital with like zero downtime we're already under pressure to say like hey doctors and nurses we're going to run out a a test like ehr maybe it's cool like that's not happening right so the certainty level you need to have that it's great is high the amount of like povs pocs you want to run before doing anything is real.

30:42The time they believe they have for this transition is longer than anybody in Silicon Valley believes. And the contracts actually allow them to think that. And I think the change management readiness on most of non-private healthcare is real. Private healthcare in the U.S. is maybe a little bit differently because they're obviously driven by some different incentives. But probably they won't start with the clinical reality. They're going to start with medical coding and billing and invoicing and all that stuff. I'd love to understand from you navigating regulation in this for you. How have you done it?

31:11What have you learned? How would you advise out of the founders to go about navigating regulation in this space? Yeah, this is obviously something we think about a lot. And first and foremost, I think that the narrative about regulation is really wrong from a startup perspective. So what I always hear is this. Oh, like Europe is behind. They're stupid. They want to regulate their way out of it. But it's not happening here in the U.S. and regulations, just get rid of it and we'll have a better healthcare market. I just think the first principles here is that healthcare is not here to use technology.

31:49They're here to care for people. They use technology to care for people. That's great. And healthcare is a last resort. Health as a wider industry has a lot of pre-hospital stuff that happens to make sure you never get to the hospital. But you end up in a clinic, even at your doctor's office, it's probably already not great. Like, you won't go there unless, like, it's not like, I'm going to sit at the doctors today and hang out. Just like the waiting room is so cozy. Like, coffee is great. Like, nobody does that. So I think the narrative of, like, oh, it's so stupid, all the regulation, I think it forgets one thing, which is when you and I as patients are in the hospital and we're lying on the bed and we're waiting to get whatever treatment, right?

32:26We don't want to be asking, like, are you guys ISO certified? Is the LLM sort of accountable? Or, like, who's accountable if it makes mistakes? Or that surgery robot, is that, like, FDA cleared? or it's like something you're testing out of China. I wouldn't want to go there. You wouldn't want to go there. Nobody wants to go there. And even worse, we both have kids. Send your kids there? No way. So the permanent thesis of healthcare has to be that whenever you're in healthcare, the actual reactive part of the system, everything has to be so you never have to question it. But if we never want to question it, any of us, when we consume it, it has to live up to a really high bar.

32:59And that bar has to be pretty standardized. And that's why we have regulation. It's not because Europe doesn't like American technology or Chinese technology, it's because we want one standard. So, you know, if you get treated in Sweden or you get treated in Denmark, we have the same thing. Or in Europe, if you're in Germany and you go to France, you know you get decent treatment. So how we're thinking a lot about this every time we look at the regulatory narrative is are we heading in a direction that will sort of make it better and easier for clinicians to feel in healthcare that they have less risk and they can do more, then it's probably going to be a permanent factor.

33:30It's probably going to stay permanently. If it's an idea for regulation that is much more at the edge or changes workflows that might not be helping doctors, it might go through a 10-year process to get into reality, right? So we tried from first principles just to look at what will make it easier for healthcare to consume risk, have a better time treating patients, and feel a clear standard for all their vendors. That's probably going to be reality. And what we're seeing right now in the U.S. even is that state-by-state AI regulation is changing. There's increasingly, even under Trump, adoption from European and especially UK AI standards.

34:03And I think the FDA and the European authorities have worked together before. They might want to do it again in the future. I don't see any path where regulation doesn't happen in healthcare. Just it has to focus on making the clinical reality better. So that's been our core focus. So in other words, to summarize, founders should reset their thinking around regulation, seeing them as first and foremost instrumental for giving us a safe healthcare system. That's one thing. Then my question to you is, how do you navigate it? What would you say have been your three biggest learnings when it comes to navigating the regulation of the healthcare system?

34:42I think first and foremost, everybody should remember, the moment you know you have to get approval from German authorities or whatever, French authorities, or I could go on, There is an entire industry that's created to help you do that. And the service part of the industry is selling, if you look at their business models and then their incentives, it's selling ours. If I'm selling you ours, I probably want to sell you a lot. That's a great business. So I'm making it pretty complex. So our learning is very much that a lot of startups, they end up raising a seed round or an A round. Some of them come to us.

35:13They're in the middle of whatever approval kind. And they have been swallowed up by a CISO who is like external and working in a company. and it's very complex, all of it. And there's an entire sort of compliance company apparatus around them saying that if you don't do it like that, you probably don't get it. And usually these companies get very slow to market and they haven't really, some of them at least, taken it upon themselves to really own the minutiae. So I think what my co-founder did very early was that he hired the entire compliance team inside Cordy. We don't use external as much as we can, if ever.

35:47We don't do that across what we're in 11 countries now. So we cover quite a lot of ground, but we do it all internally. So I think our compliance team is bigger than our HR team. And our goal here is not either or is bigger. It's just to deliver what we need to deliver. And we're trying to productize it more and more. So even our customers, some of our biggest EHRs, they come to us when they have to go into a new country and like, do you guys, have you built a stack we can build on? So now we're increasingly turning it into something where, yeah, like use our entire stack and we'll help you work with the government, right?

36:14So I think as we've, my advice would just be, I know it sounds daunting. It is really not that daunting. The moment you get out of the advisory industry that wants to tell you it's the hardest thing in the world, it's definitely much easier than building your company. It's actually taking it upon yourself to own it as much as you can and find the path of least resistance, and it's probably the path with least amount of people involved. It's actually a surprisingly simple, though I imagine for many, counterintuitive learning there. but you would always say that you would want to control and own the success in any space that's absolutely instrumental to you.

36:55And we always refer to healthcare as a heavily regulated industry, and that's why you will not succeed as a founder in it. That's been the spiel against it. But obviously, given that being the constraint, you want to own it then. It makes a ton of sense. Andres, let me ask you a different question, and I'm going to go super wide now. and instead of focusing on Cordy so much, then focus on your learning from raising an inventor and being an inventor for a long time. You are, of course, not only the founder of Cordy, you've got past successes behind you. You've raised hundreds of millions. I would love to ask you if you were to speak bluntly to our founder audience about how they should be thinking about navigating venture, what would be your biggest things that you would say to them?

37:41I think the advice I give the most these days, and I think founders are less excused than ever before is the sort of the canonical get to product market fit. I think everybody knows it, but like, what does it mean? I think what I say the most, like get to actual product market fit. And I don't think there's more or less any excuse for not being capable of getting a really nice POV to some level of quality where you can get that kind of feedback. There's ample amount of VCs available in the world and they would definitely invest in you, but I think the bar has just increased so much. And I think that's fair.

38:13and maybe only good. So if you look at the macros as I see them, more money is going to the winners. They accumulate faster. And it's getting harder and harder for late people getting to A and B round because ultimately you need so much more firepower to be able to. And I think the sort of daunting challenge for many is like, oh, I need to quit my job. I need to make all these sacrifices. And I think the navigating of VC is like, VC should never be sort of your jump parachute. It doesn't feel like, at least the founders I know, that's not the ones who have the most success who like needed that to jump.

38:41Build for the jump. And the moment you feel that your parachute will hold, you want wind in your parachute, like that's when you get VC money. I think navigating that, like there's a lot more pressure on founders today. They should build a lot more. They should get a lot further. And the best is a global problem, right? When I raised my first round in 17, 16, whatever, there was much fewer people building. They were so formidable. The best were great. But now I'm competing with people who can place money elsewhere. They have much more access, much more visibility on deals. And founders come from all over the world and they're formidable.

39:09So I think the local problem of like being a local winner in Scandinavia that's like great but like you're in a global play or you're not for VC money and you're against a global horde of companies or you're not. How about, you know, we shared off the line, we talked about how some VCs behave. Would you care to share a bit about how you would recommend founders or advise founders to manage VCs and what's okay behavior, what's not okay behavior, what's just not okay but you've got to just accept it because this is the name of the game. It's a critical service industry if you're building in deep tech.

39:45You absolutely need the support of fantastic VCs. And I think from my own experience, we've been very lucky to have some fantastic ones. I think our last round was led by Atomicon Process. And I think there's not a week where I don't pester Atomicon for something and they're fantastic. So I think first things first, I think it's much more about being very thesis driven as a founder, which is a luxury you only get if you get to a certain level of success. But that's why you need to build for it. Because I think the vacuum between the median VC and the best VCs are widening. and you want the best because they unlock capital, a different kind of growth journey, different kind of exit scenarios, whereas I think the media and VC do not.

40:20So I think the first thing first, I think just founders need to make every problem, including building VC relationships, founding, funding, and leading a board and a group of VCs, they need to make it a product and a founder and a builder's problem because ultimately capital allocation obviously is the name of the game and you're trying to allocate other people's money as best you can. So I think generally, especially in Europe, there is a lot to be learned still on how to play a game with international money allocators at scale and living in that kind of pressure and being hyper-aggressive as a founder and CEO, which I think many European startups still are learning.

40:54But I think there's a big vacuum in sort of what we do and how we work in Europe versus the US. It's not increasing at all. Like the best companies in Europe are fantastic today and they're doing it more and more. But I think generally there is a lot to do And I still think there's also a lot to learn from European VCs on what sort of best in class kind of behavior. If I should comment on that, which I think I must, what I would say is that you're absolutely right about your point about the best and the non-best, so to say. I agree that the, I don't know if I agree that the chasm is increasing.

41:30I actually maybe think, meaning between the good and the less good, I maybe actually think that it is not. It definitely is in terms of being able to raise large funds and thus stay competitive. But I think that behavior in Europe from VCs towards founders is increasing for the better. There's just a difference between who will be successful in fundraising and create the funds that are off a size where it's truly relevant. So I would actually say that we're in a better position from that perspective. But I think you're absolutely right about the importance of optimizing for the best names and the people that really know what you're doing.

42:11And I'm definitely seeing that that is getting harder to diligence because you need to not only diligence whether the person on the other side feels strong and good, but you also got to diligence whether this fund is on a trajectory that is good because that will, as you said, are building in a deep tech space. You need someone who can fund you quite a long way because you're going to need that money later. And don't you think like when looking at sort of the leading funds in the U.S., let's take healthcare. The leading funds in healthcare, like a general catalyst, they're calling themselves a platform, which in my universe is a P kind of term.

42:52It's not a VC term. But all of a sudden you're trying to blend categories, which I think is really smart. But I see Andreessen do the same thing. They're also blending, talking about platforms. Like general catalysts bought a hospital. Like if you mean healthcare and you have a thesis, like that's ultimately condition, right? Roll out all the VC investments you've done in your own hospital. So my point being, I think we're catching up to where our VCs in the US were. I think the VC category in the US is moving as well. And I increasingly just think that there is a lot of power law going on when you can take a lot of capital, maybe even more, maybe marry it with GPUs.

43:24I think that's something we've been pitching to many VCs in Europe, like what's your GPU strategy? We've done that for years. Most VCs in Europe, especially the big ones, still don't have one. Massive mistake in my point of view. If they've done it earlier, they could have been somewhere else. Elaborate on that. The GPUs. So if you give a pre-seed company$3 million to build a vertical model for generating ad spend, or you know you're going to spend a dozen of dollars on inference with one of the big models, but they need to build something, tune something, build their own, and they need to go to NVIDIA and buy it themselves.

43:55I think the VC perspective in the US has been like, we're just going to buy GPUs ourselves. Then we're going to start distributing. Like Andreessen have done that. Many others have as well. I think Ned Friedman is, I think, a leader there. Their VC, like angel fund turned VC fund. They had Andromeda, one of the biggest VC angels, slash angel clusters in the world of GPUs. They ran themselves, gave it to their investments alongside money. Obviously, that kind of thing on scale, I think. Ultimately, if you control the electricity and the tokens for the generation of AI startups to come, you have a completely different offering than just sort of passive money and the argument about, oh, we're like, we're here to help.

44:28But I think sort of a lot of VCs forget that every other VC says that to a pre-sie company. We're here to help. We have network. We work with founders that build great companies. But everybody do that, right? And I can find the same founder as a venture partner in multiple funds. So I think ultimately, as we head into a new paradigm, the narrative, the category blend in the U.S. feels real. The accumulation of money is bigger than ever. They're betting much harder on their winners. They come with other things than money, like GPUs, and they're embedding themselves much more into ecosystems in the market they play in, like healthcare.

44:58The last point is, of course, one that's very strong around verticalization of VC and specialization of VC, however you want to frame it. I think it's incredibly interesting to hear you talk about the GPU strategy and the importance of that. I completely agree it's not something I've seen in any way being led by a lot of European VCs. And then I think that you're absolutely right when you're also talking about the importance of this merging of asset classes. It might sound to the typical founder listening to this, why are you talking about this? But one of the really big problems that we have is this stage of going from venture-funded to private equity-funded.

45:37And that's, of course, because most founders will, most especially deep tech founders, will relate to the problem of not being able to go from the university setting or the grant-funded setting to raising your first private money. you have the exact same problem happening at the end of the stage, at the end of the journey where you're getting from venture into private equity, where a venture still underwrites you as a venture case and loves you for that, but private equity is finding it very difficult underwriting you as that. And for that reason, these platforms that are operating across that spectrum are very interesting to get funded by.

46:13Yeah, and I think we had sort of this European narrative a couple of years back when we were still doing like a pre-A round. I remember there was this period, you might recall, where everybody was like, oh, we're going to build a seed fund. Everybody had a$100 million seed fund all of a sudden. And it was like, there was no money for seed, which in my point of view always felt a little bit weird seeing it from a founder's perspective. So obviously, there was a lot of companies that needed funding. Totally agree. And there was ecosystems in Europe like Vilnius, great city, fantastic town, had less access.

46:42Totally true. But that was also true in the US, right? There was also a time when Boulder wasn't attracting as much capital. Austin still felt like a weird place in Texas for most people. But now it's just Austin and it's awesome. But that's also the reality for Europe. There was a maturing of sort of the ecosystems. But I think everybody was investing in seed. And now everybody's saying the different thing, which is like, what do we do at growth? What's the ABC realm? What's the platform beyond? How do we build a platform into public markets? I think we have a really big problem with it's almost impossible to have a meaningful IPO in most European countries at this point.

47:12so there is no path there that means investment banking quality and talent is less and obviously we have some great ones in copenhagen but there is a massive gap here that copenhagen maybe cannot continue to attract the best investment bankers in the world to help people go public on a local nasdaq less people than do that means the next generation doesn't have any references of that was a great going ipo or dual ipo in copenhagen so they go to london they go to us and ultimately that's I think the trajectory of like there is so much pressure that because the top part of the growth funnel doesn't work very well and there's no path to local IPOs and I think that's a massive hindrance for local ecosystems and if we don't plug it really fast and finding meaningful ways of helping people to scale beyond just we have great advice then I think like ecosystems will cap out and more people will continue to move to London and especially US.

48:01How do you advise fellow founders about going to the US because it's obviously one thing many have to do for the commercial side, some do it for the fundraising side. First, maybe share, why did you do it? Why did you choose to move the commercial headquarters to New York and yourself as well? And what has it unlocked for you? In 1617, when we were creating what became Cordy, we were already in San Francisco. So we actually moved back to Europe for talent and for data access and due to local partnerships with healthcare. So we had actually not planned on building from Europe, but ended up doing it.

48:36And we've been very happy we did for many reasons. We then, or we've always had a presence in the US. And during our time, the vast majority of our customers have been American. But being on the West Coast versus being on the East Coast, massive difference in time. So when we started the New York office, it felt meaningful we were there. Our two biggest customers are American still. So we wanted to be there. I wanted to scale that, build a team out there. We got fantastic leadership now. So that was sort of our case. And obviously, we already always planned on working out of the US. And both of us love it.

49:08And still very passionate about Europe and Denmark for that matter. But that was our plan. And I think what I say to people is obviously, like, I speak a lot with seed levels, like seed stage founders, A-round founders who are like told by their seed stage VCs, like, go to the US, you have to go to the US. And I think, A, I think the sort of the SaaS logic of the European market being too temp constrained, I think is not true. I think in our space, we have a fantastic startup up in Sweden called Tandem and Healthcare. They're doing great across Europe, and they have no plans of doing US and all power to them.

49:36They're building a fantastic company. I could go on. There's quite a few. That doesn't mean you shouldn't go to US, just like the VC fantasy of everything is easier in the US, and there's a much bigger TAM that will help you a lot. True, but tough. The cost is also much higher. The talent acquisition is much higher. You don't have any network there to help you talent acquire and keeping them happy. So I think post-proto market fit with real scale and sort of a platform to open new markets makes sense. Or if your cornerstone customers like ours are there, it makes sense. But I think rushing to the U.S.

50:04today, like 26 when Europe is awakening, I'm not sure it's the best for everyone. Let's land on that sentence. I love that. Andres, thank you so much for joining me on the podcast today. We did an episode earlier that was about an announcement of yours. Now we did the Bigger Deep Dive. Thank you so much for that. Thank you sir, it's great having you.

From the publisher

In vertical AI, choosing a narrower market can look like giving up scale, but for Corti it created the focus needed to go much deeper into healthcare and build around the accuracy, compliance and real-world complexity that clinical environments demand.

Andreas Cleve, Co-Founder and CEO of Corti, joins us to explain the thinking behind that choice and what it has meant in practice.

Andreas describes healthcare not as one market but as many highly specialised ones, each with its own workflows, terminology and requirements.

That depth means bigger general-purpose models alone are not enough. Corti has instead had to build further down the stack and take greater control over the technology and infrastructure required to get AI into production.

The trade-off becomes especially clear around accuracy. AI can perform well in a demo and still fall short in a clinical environment, where one incorrect word or detail can change the meaning entirely.

Andreas explains why clinical-grade performance depends on much more than training better models, from working closely with users to adapting to individual specialties and solving the last-mile problems that appear when technology meets real workflows.

The conversation also explores why Corti brought compliance in-house rather than treating regulation as something to outsource.

For founders building in healthcare, legal, financial services or other high-stakes domains, the broader question is how to decide which constraints to accept, which capabilities to own and where going deeper can create an advantage that broader competitors struggle to replicate.

Highlights

  • Why Corti chose vertical AI over a broader horizontal approach
  • What founders give up when they deliberately narrow their market
  • Why bigger models do not automatically solve regulated-domain problems
  • Why 90% accuracy can still fall far short in clinical environments
  • How specialisation pushes vertical AI companies deeper into the stack
  • Why getting AI into production requires more than model performance
  • Why Corti built its compliance capability in-house
  • How founders can think differently about regulation and other high-stakes constraints


Timestamps

  • (00:00) Intro
  • (01:45) What Corti is building
  • (03:45) Why healthcare needs specialised AI
  • (05:45) Why bigger models do not solve healthcare
  • (07:40) Healthcare is not one market
  • (09:50) The trade-offs of going vertical
  • (11:50) Funding a capital-intensive AI company from Europe
  • (13:15) Why Corti is doubling down on Copenhagen
  • (16:45) Copenhagen’s talent problem
  • (19:30) Why 90% accuracy is not enough in healthcare
  • (23:00) Solving the last mile of healthcare AI
  • (25:45) Where AI adoption is happening fastest in healthcare
  • (31:20) Why regulation is not the enemy
  • (34:40) How founders should approach compliance
  • (37:40) What founders should understand about venture funding
  • (42:30) Why European VCs need a GPU strategy
  • (46:40) Europe’s growth funding gap
  • (48:00) When should European founders move to the US?

More from EUVC

All episodes