Unicorn University Student Jáchym Vrtiška: Building the AI Startup HealthyNess While Studying

Let’s be honest, there are definitely moments when it’s really demanding. I’m basically working 24/7, so one of the big questions has always been whether I can manage that alongside university.

Honestly, I wouldn’t be able to do it at a regular university — not so much because of the academic difficulty, but mainly because of mandatory classes. At Unicorn, I was able to arrange my schedule to fit my needs, so I have classes either in the morning or in the evening. Attendance is not mandatory, so if I miss one from time to time, it’s not a big problem. The study materials are always available in the course portal, and the lecturers are quite understanding.

That said, I do normally attend classes. Since I’m also a co-founder, I can structure my work the way I need to, which definitely helps too — at least on paper. In reality, unfortunately, I still work 24/7. 😊

Like any proper startup, we’ve already been through at least three pivots. At first, we wanted to build an app for end users. Then we significantly reworked the whole concept, and today we mainly focus on a B2B API and our own AI engine.

The startup really began when my co-founder, Matyáš Hronek, called me and said he wanted to launch a startup focused on an application that would take patients’ medical history. At that time, I already had several clients in AI and technology, but the idea immediately made sense to me. I thought, this is exactly the kind of thing that has real potential to change healthcare.

The first impulse was actually quite simple. Matyáš’s dad was using an English-learning app where he communicated with an AI avatar, and that’s when he thought, “This would be great if, instead of a teacher, it worked as a doctor.” That was the starting point from which the whole idea gradually evolved into its current form, which is now quite radically different.

The main problem we want to solve with AI is the error rate in healthcare. We believe that a large share of these errors already arises at the stage of collecting information from the patient. That is why we are trying to create the right structure for medical history intake and then enable clinics to work with that data effectively in whatever way they need. In that sense, our platform is an all-in-one solution.

As a result, both prevention and care itself can become more accurate, faster, and more efficient. At the same time, we want healthcare to be more accessible and significantly less expensive than it is today.

This is extremely important for us because my background is primarily technical, while Matyáš Hronek’s is in marketing. The good thing is that both of us are also fairly well oriented in business and sales.

At the same time, neither of us is a doctor or comes directly from a medical background. That’s exactly why we knew from the very beginning that if we wanted to build something like this properly, we could not do it without experts.

Today, we have a team of eleven people, four of whom are doctors or final-year medical students. For us, that is absolutely crucial, because defining the logic of the whole system is not just a technological challenge. You need to be able to translate medical procedures, clinical thinking, and the experience of real doctors into a meaningful and functional structure that AI can then work with. Above all, it also has to be clear where the sources we use to build our system come from. That’s where the medical students are incredibly valuable — they actually prepare studies and papers for us based on validated sources.

That is why, from the very beginning, we have been developing everything together with our medical board. Step by step, we have worked together to determine how the whole system should function, how medical history intake should be conducted properly, and how the other features of our platform should work.

And why do we think it is so important to work with experts? Because no one ever knows everything. It would be naive to think that someone can truly understand medicine solely on the basis of online sources or articles. Medicine is one of the most complex fields there is, built on decades of experience, practice, and clinical reasoning.

So when we have the opportunity to work with truly high-quality doctors and medical students directly within our team, it would be a shame not to make full use of it.

When I started my second semester, we had a course focused on the fundamentals of artificial intelligence. And even though at the time I kind of thought I already knew everything — obviously exaggerating a bit — that course really opened my eyes. The basics of Python programming and the Computer Systems course we had in the first year also helped me a lot. For me, those were genuinely great subjects with real-life use cases.

I’ve never seen education as something where you just sit down and mechanically memorize theory. I have a hard time learning things when I can’t see any real application for them. Of course, not everything you learn at university is something you will use directly, but for me it is important to see where that knowledge makes sense in practice.

And that is exactly what often happens here. I’d say that around 60% of what we learn is something I also come across in real projects, which feels like a lot to me by university standards.

At the same time, I really appreciate the lecturers’ approach. Whenever I ran into a problem and needed help from someone more experienced, I could simply go to the teachers and talk the whole issue through with them. The environment is very open and human, the teachers know students almost by name, and they are very proactive.

That individual approach is probably what I value most about the university.

Honestly, this is a very difficult question, because predicting the development of AI more than half a year ahead is almost impossible. So this should definitely be taken with a grain of salt, but I do think a few directions are already visible.

The first big thing will be the wider local use of AI. I do not necessarily mean working with sensitive data, but rather the smart integration of information based on how a person moves around their device, where they store their data, and how they work with it. We can already see some early signs of this, but the real potential is still on a completely different level.

The second area is cybersecurity. I believe that in the future, benchmarks and testing focused specifically on cybersecurity will become much more important for AI than general performance benchmarks alone.

At the same time, I expect AI development to become much more heavily regulated by governments. Model training, data handling, and system deployment will all be subject to greater oversight. In my view, governments will try to be more proactive in this area, which may be beneficial in some respects, but in others it could slow development down quite a lot.

As for whether AI is a bubble, I partly understand that argument. I think a lot of pseudo-startups, typically various “fancy to-do lists” with an extra AI feature, will gradually disappear. Especially if that AI feature is just a marketing add-on and does not actually deliver real value.

This is also a very important point when starting a startup: artificial intelligence is not always necessary. That is something we try to stick to at HealthyNess as well. We do not put AI where AI does not need to be. It is more expensive, more complex, and often riskier.

On the other hand, AI is obviously a huge opportunity. If you know what you are doing, have a clear use case, and it makes sense to integrate it, then go for it. But one thing really does apply: research, research, research.

I think a lot of people in AI still underestimate one fundamental thing: these models are not all-knowing — they are simply very good at predicting the next word or character. You are the one who has to provide the right context and direction. People often expect that they can briefly describe something to AI and a perfect result will “magically” appear. But that is not how it actually works.

The reality is more that you need to have a well-thought-out plan and a clear idea of what you want to achieve. AI can then significantly accelerate execution, help with iteration, or even uncover mistakes. But the core direction still has to come from a human.

What continues to surprise me is how much people underestimate one very basic thing: using the right models for the right tasks. And I understand that for an average user, switching between different models depending on the task is not always easy in practice. But it is important.

If you need to generate images, use an image model. If you are writing code, use a model optimized for programming. If you are handling emails, analytics, or text work, there are again more suitable models. The difference in results can be huge.

Another thing that I think people still underestimate is working with context. There are studies showing that as the amount of context increases, models begin to degrade quite significantly in the quality of their outputs. I’m personally a big fan of Anthropic and their models, and even there you can see that once the context window gets filled beyond a certain point, the quality of responses starts to noticeably decline.

That is why it is extremely important to keep communication concise and get to the point as quickly as possible. If you can solve the problem within the first half of the available context, do it and free up the rest of the context.

The same applies to multi-agent systems. Everyone talks about them today, but people often underestimate the reality there as well. Multi-agent systems consume context very quickly, they are expensive, and most importantly, they are useless if the individual agents cannot communicate effectively and share information with one another.

In healthcare, the whole issue is much more sensitive, because you have to deal with a huge number of regulations and security measures. We plan to implement many of these mechanisms, but at the same time there is also one major technological and ethical challenge: the moment you are dealing with a user’s health condition, you bear a certain degree of responsibility for how they use that information.

Even though we have it contractually defined so that responsibility for the patient’s final decision lies with the clinic or healthcare provider, I still believe there is a certain moral responsibility to make sure that a patient does not make the wrong decision based solely on information they received from HealthyNess or through systems built on our API.

That is why I would never underestimate communication toward the user. It is important to keep emphasizing that the outputs are not a final medical conclusion or a substitute for a professional examination. They are supportive information and a way to help both the clinic and the patient better navigate the given issue.

As for the ethical side of AI in healthcare, it is actually fairly easy to understand, but very difficult to set up correctly. You have to realize that these systems are already not that far from a situation in which they begin to partially substitute the work of a doctor.

Of course, we always say that we are not replacing doctors, and that is an important disclaimer. On the other hand, we would not be entirely honest if we claimed that, with a truly high-quality system, a patient could never make certain decisions autonomously in the future based on AI recommendations. For example, they might directly book a specific examination or decide to avoid visiting a doctor altogether.

That is exactly where the biggest ethical conflict arises. We are often asked, “Aren’t you already replacing doctors?” We do not think we are replacing them, but at the same time we understand why it may seem that way. In some scenarios, that boundary can feel extremely sensitive.

I would definitely recommend that students start by doing research and finding a topic they have some personal connection to. Part of my family, for example, works in healthcare, so this field was not completely unfamiliar to me from the beginning. I think having a personal connection to the problem you want to solve is extremely important.

At the same time, I would say one important thing: you have time. At the beginning, I often felt like I was under enormous time pressure and had to achieve everything as quickly as possible. Of course, that has its downsides — for example, I already have a receding hairline and back pain. But in the end, it really does not matter whether you are 40 or 17 when you start building something.

A lot of people say, “just start,” and that is true, but it is not a good idea to jump into things completely blindly. If you are going to jump into something, it is good to at least check where you are going to land. Either you need to understand the field you are entering really well, or you need to have a clear idea of the product you want to build and only then look for its application. But you always need to have at least one of those two things firmly in hand.

For me, the absolute foundation is to have a general understanding of how artificial intelligence works. I think the most successful people in AI in the future will not just be those with extremely deep expertise in one narrow area, but also those who have a very good horizontal overview across technologies and disciplines.

That means you can be an expert in chatbots or AI products, for example, but at the same time you should also have at least a basic understanding of databases, infrastructure, security, working with models, or backend systems. You do not have to be the best database architect in the world, but when someone explains why you need a certain database or architecture, you should understand what that means, at least on an entry-level level.

And one more thing — the founder of Palantir once said that the best people in the age of AI are those with ADHD. I have ADHD, my colleague does too, so hopefully he is right.

Being a company mentioned in Forbes is, of course, a great feeling, but for us it is mainly proof that we have truly exceptional people on our team.

More specifically, it was an interview with Martin Ouzký, who is the head of our medical board. For us, that was important precisely because Forbes chose him for his expertise and the projects he is involved in, one of which is HealthyNess.

Media attention means several things to us. Of course, it is a big motivational factor, and at the same time it helps raise awareness of what we do. But above all, it is a certain form of validation. It shows that what we are building makes sense and that the way we think about the problem resonates beyond our own team.

If someone looked at our product and thought the whole thing made no sense, opportunities like this probably would not come at all. The fact that we got there meant a lot to me personally. I had always seen it as a kind of milestone — something I would want to experience one day. I just did not expect it to happen already at the age of 20, and through someone from our own team. That was truly incredible.

At the same time, I also believe that media attention should never be the main goal. The most important thing is to have a product that genuinely brings value to people and solves a real problem. Everything else comes only after that.