AI Engineer · Product & Operations · Berlin
I turn repeated expert work and fragmented inputs into safe, inspectable self-service workflows — with evidence, failure behaviour and human authority made explicit.
Before engineering, I ran e-commerce and service operations. That is why I tend to start with the bottleneck, not the model.
Portfolio → · Web CV → · LinkedIn
Operations self-service · bounded agents · human release
A reusable agent runtime where permissions, required evidence and consequential actions are controlled outside the model. Public HausPilot cases show the full path from incoming request → prepared work → human release, including fail-closed behaviour when required records are missing.
Try the workflow →
Document self-service · evidence → resolution → authority
Turns document packs and versioned rules into reviewable findings, exact next actions and prepared hand-offs. The public workflow shows missing-evidence, policy-exception and clean-pass cases — including what happens after a problem is found.
Try the workflow →
Source-backed investigations · provenance · uncertainty
Question-first investigations across public records. SafeTrace keeps evidence, inference, uncertainty and the next research move separate, with original sources inspectable behind consequential claims.
Explore SafeTrace →
⚖️ GitLaw
Source-grounded legal research · retrieval · citations
Search and inspect German federal law with traceable sources. Retrieval failures can become regression cases instead of disappearing into a generic answer.
Try GitLaw →
repeated request → map sources + authority → build the smallest useful workflow → make evidence visible → fail closed → measure → earn autonomy
Python · FastAPI · JavaScript / Node.js · TypeScript · React · SQL / Postgres · LLMs · RAG · Agents · MCP · structured outputs · evaluations
The recurring engineering principle is simple: models can interpret and propose; evidence, permissions and authority should not live only inside the model.
CareOS · TrustReady · SafeVoice · Citizen Agents · Council
Looking for a strong team where I can own useful AI problems end to end.