I build things when the problem is still messy.
Most of what interests me sits somewhere between AI, automation, product thinking and real-world workflows. I like taking a vague idea, breaking it into something testable, building the first useful version and then trying to prove where it works, where it fails and what should happen next.
I use AI heavily in development, but I’m not interested in pretending the model did the thinking for me. The useful part is choosing the problem, shaping the workflow, spotting weak assumptions, testing edge cases and turning the result into something another person can actually use.
- AI workflows and agentic systems
- evaluation, regression testing and release decisions
- automation and internal tools
- product UX for technical systems
- the awkward edge cases that appear right after someone says “it works”
A working reference MVP for checking AI changes before production. It compares a current and new model / prompt / agent workflow on the same cases and turns quality, safety, cost and latency evidence into a clear release decision.
Built with FastAPI, PostgreSQL, React/TypeScript, Docker and Playwright. The public version uses synthetic data and explicit trust boundaries rather than pretending to be a production certification platform.
My background is in customer communication, marketing and operations, which turns out to be surprisingly useful when building AI systems for humans. I care about understanding what the person actually needs, where the failure would hurt, and whether the thing we built solves the right problem rather than merely looking clever in a demo.
Based in Prague. Working in Czech and English.