AI Engineer Markus Kruth talks about an ongoing project in which Data Design is helping build an AI-based workplace wellbeing coach. The project brings together language models, personalization, research knowledge and the handling of sensitive user data. The technical challenge is not just getting AI to answer questions, but building a coach that remembers, adapts and supports the user over the long term.
Who are you and what do you do?
I'm Markus Kruth and I work at Data Design as an AI Engineer. My work centres on software development and on implementing various AI systems in practice. My core expertise is turning data and AI algorithms into durable software solutions whose maintainability and performance hold up against real-world demands.
Has any project stayed with you in particular?
The one that has stayed with me is a project still under way, in which we're developing an app for workplace wellbeing coaching. It has been really interesting to get to design an entirely new product from the very beginning, and to see how different perspectives shape the outcome.
Marketing is interested in different things about the product than the UX designers are, and technical development has its own emphases entirely. In the end our job is to turn the idea into a working product: to build the app and the features that create the actual user experience.
What is the project about?
The starting point is supporting wellbeing and performance at work. The signs of exhaustion can build up for a long time before a person stops to notice them.
We're answering that need by building a digital wellbeing coach that helps the user recognise the early signs of strain and offers concrete ways to look after their own wellbeing.
At the same time, the coach has to adapt to the user's individual goals. One person may want to focus on recovery, while another wants to become a better leader, for example. The emphasis of the coaching shifts according to those needs, but the same main goals remain underneath: supporting wellbeing and performance at work.
What kind of solution are you building?
The core of the solution is the AI coach. The hardest part is that the coach has to satisfy several requirements at the same time.
It has to be able to take account of changes in the user's situation and adapt to their goals. It also has to remember earlier conversations, because without continuity it isn't really coaching but answering individual questions. At the same time, the answers have to be grounded in trustworthy information and scientific research.
One of the most interesting requirements concerns the interaction itself. The coach should feel like someone you're talking to, while it is at the same time essential for the user to know that there isn't a real person at the other end. In user research, this in particular emerged as one important need.
The technically most interesting work comes from reconciling these requirements: how do you build a coach that is personal, consistent and trustworthy.
What challenges have come up in the project?
This was Data Design's first mobile app project, so at the start we also had to work out the practices of mobile development and everything that has to be taken into account in it.
Fairly quickly, though, it turned out that much of it is close to the web development we already know. The threshold proved lower than we first thought, and at the same time the project has built expertise we can use in our next mobile projects.
How have security and data protection been addressed?
When you're handling information related to wellbeing and health, data protection is not something you can add to the product afterwards. It has to be taken into account in the architecture from the very start.
The basic principle is data minimization: information is processed only where and to the extent that running the service requires. Handling personal data naturally also means taking the requirements of GDPR into account.
On top of that, you have to design carefully how information is transferred between the app and the backend systems, how it is protected, and at which stages identifying data is needed at all. The more sensitive the data, the more important it is to minimize both the amount of information processed and any unnecessary exposure.
What has been the most important insight of the project?
When a project involves experts from several different companies and areas of expertise, a shared vision can easily start to come apart and mean slightly different things to different people. In that situation a concrete proof of concept is a surprisingly effective way to get everyone looking at the same thing again.
The turning point in the project was indeed the first demo. When the team saw for the first time how the coach actually responds to a user, the conversation immediately became more concrete. Instead of talking abstractly about what the coach should be like, people could say: "It should react differently to that" or "In this situation it should ask about this."
So the demo wasn't just a technical milestone. It was the tool through which a shared view of the product began to take shape.
What has the reception been like?
So far the reception has been good. The product owner has been actively involved in the development and has continuously brought views on what the coach should and shouldn't do.
We're still in the pilot phase, so it's too early to draw final conclusions. Based on the results so far, though, the direction looks promising.
Could the same be applied elsewhere?
The app itself has been built for a specific purpose, but the expertise the project has produced is far more broadly applicable.
One concrete lesson concerns mobile development. Now that the first mobile app has been built, starting the next project is considerably easier.
An even more significant lesson concerns the AI coach. We've learned a great deal about how a language model can be steered towards a particular purpose, how memory and context are built for it, and how its behaviour can be bounded and developed.
The same questions come up in many other AI products. That's why the expertise the project produced transfers much more widely than to this one app.
What has been the best part of the project?
The cross-disciplinary design. When developers, UX designers, marketing people and other experts all look at the same product, things surface that one group alone might not notice at all.
As a developer you easily focus on how to make a feature work. Someone else asks whether the user understands it, whether they need it, or how it should be talked about. Combining exactly these perspectives is what makes the product better.
How does this continue from here?
The main focus now is on developing the coach itself. During the pilot we've identified situations and patterns of behaviour where we want to improve how it works, and we're taking those into implementation systematically.
The goal isn't just to make the AI a better conversationalist. The goal is to make it a better coach: one that understands the context of the conversation, adapts to the user's goals and can offer the right kind of support in the right situation.
