Projects
Three core LLM patterns, end-to-end built
Retrieval (RAG), agents and structured extraction — each tested, evaluated, dockerised and GDPR-compliant. A shared Python foundation, three different domains.
RAG · Retrieval In progress
PflegeLotse
Care law is complex — and answers must be verifiable
Python 3.12FastAPIpgvector
Eval: Recall@k · Faithfulness · Abstention rate Details → Agents · Tool-calling In progress
PostLotse
Email overload costs SMEs hours every week
Python 3.12FastAPIHTMX
Eval: Routing F1 · Tool-call success · Escalation rate Details → Structured extraction In progress
BelegLotse
Typing in receipts is slow and error-prone
Python 3.12FastAPIOCR
Eval: Field precision · Field recall · Validation rate Details → Shared foundation
One skeleton, three projects
Clean Architecture, tested (pytest), CI (ruff/mypy), Docker and GDPR documentation — built once, reused across all three projects.
Python 3.12FastAPIJinja2 + HTMXPostgreSQLDockerGitHub ActionsMistral / OllamaClean Architecture