AI Engineer · Product Builder · Berlin / remote
I build AI systems that turn complex, high-stakes work into clear, reviewable action — grounded in evidence, explicit about uncertainty and designed to keep humans in control.
Portfolio → · CV · LinkedIn · Culture
| Project | What it proves | Try / inspect |
|---|---|---|
| ⚖️ GitLaw | Source-grounded legal AI across 5,936 German federal laws: hybrid retrieval, citation verification and reviewable next steps | Try GitLaw → |
| 🏥 CareOS | Clinician-first clinical context architecture: source-linked patient truth, FHIR/ISiK, bounded agents, safety gates and adversarial testing | Explore CareOS → |
| 🛡️ SafeVoice | Evidence-first AI for digital harassment: structured evidence, legal context, integrity checks and human-reviewed case preparation | Open beta → |
| 📄 PrüfPilot | Public-sector Document AI: documents + versioned rules → evidence gaps, checks and reviewable next actions | Try demo → |
| 🤖 Digital Worker Factory | Trustworthy AI workers with explicit tools, evidence contracts, approval gates, audit and replayable failure tests | Open demo → |
| ✦ Council | Evidence-first multi-agent decisions with specialist disagreement, critic challenge and falsifiable recommendations | Inspect source → |
human problem → useful workflow → grounded AI → visible uncertainty → human control → feedback → regression tests
Python · FastAPI · TypeScript · React · LLMs · RAG · Agents · MCP · structured outputs · evaluations
Across the projects, the recurring engineering idea is simple: models can interpret and propose; authority, evidence and safety boundaries should live outside the model.
I’m especially interested in health, law, public services and operational work — places where good software can remove friction without quietly taking responsibility away from people.
judge-mcp · Citizen Agents · Human Systems Lab · Atelier Engine
Looking for a strong team where I can own difficult problems end to end.