Across legal research, public services and agent workflows, I turn complicated problems into practical products people can understand, test and use.
Bureaucracy that takes weeks. Rights that are hard to understand. Important information scattered across institutions. Software that creates work instead of removing it.
My instinct is usually the same: Why does this have to be so complicated? What would the better version look like?
So I build a small version of the answer, put it in front of reality, and learn where it breaks.
Each has a different maturity level. I would rather show the real state than make a prototype sound like a finished company.
Why it exists Rights are much less useful when understanding them requires legal expertise.
Find the law behind a real-life problem without hiding the source trail.
Plain-language questions → relevant German federal law → inspectable source passages → visible uncertainty. Real user-found retrieval failures are turned into regression tests.
Why it exists People should not have to manually monitor institutions just to notice when a law, benefit, budget or ruling becomes relevant to them.
Open-source watchdog agents that turn public changes into cited, reviewable briefings.
A growing fleet watches German and EU public sources, writes readable digests and machine-readable logs, and keeps a human review step before publication. The fleet is still being expanded and deployed.
Why it exists AI should remove repetitive work without quietly removing accountability.
Make AI workers show evidence, expose failure and ask before consequential actions.
The current reliability suite is synthetic. The next meaningful proof is one real office workflow running in shadow mode with human corrections measured.
Why it exists Skilled reviewers should spend their attention on judgement and exceptions, not reconstructing information from documents.
Help reviewers see what a document proves, what is missing and what needs a human decision.
Synthetic cases demonstrate the workflow; the document-intake core is working. The next proof is an institution or domain expert using their own documents.
Turn rules, documents and messy workflows into something a person can actually reason about.
Enough product to let someone complete the real task — not just watch a polished demo.
Especially when a wrong answer could affect rights, money, safety or consequential decisions.
User-found failures should become tests, product changes and measurable evidence.
The projects look different, but each begins with a human question: what friction can disappear, what becomes understandable, or what becomes more joyful?
Rights and public services only create value when people can actually find, understand and use them.
Automation should reduce repetitive work without hiding uncertainty, evidence or responsibility.
Technology can help people understand themselves and one another, not only optimise their output.
Not everything worth building needs to save time. Some things should create joy, connection and wonder.
A good product earns attention by being useful, delightful or valuable enough that someone wants to return.
I do not start with the model. I start with the person, the problem and what a good outcome should feel like.
Who is this for? What is hard today? What would make the experience genuinely better?
What do we know, what is missing, which rules matter and what should remain a human decision?
Connect data, AI and interface into the smallest version someone can actually use.
Put it in front of real and simulated cases, capture failures and turn corrections into better behaviour.
Technology should return human attention to things worthy of being human.
A parent should not spend it deciphering benefits. A caseworker should not spend it copying between documents. A nurse should not spend it reconstructing information a system already knows. And sometimes, attention should simply be free for play.
I’m looking for AI Engineer / Applied AI roles where product judgement matters as much as making the system work.
Build environments where people can understand, act and thrive.
I’m an AI engineer with a founder and operations background. I like turning messy systems into clear tools, shipping quickly, and then finding out where reality disagrees with the demo.
I care most about work that reduces friction, increases understanding and gives people more room to make good decisions. I also build games and small product experiments because usefulness is not the only thing that matters — delight matters too.
If you’re building something difficult, useful and genuinely meant for people, I’d like to hear about it.