Philip Vana
AI Engineer · Measured, sovereign-capable RAG · Berlin, Germany
Background
I'm a data engineer at Deutsche Rentenversicherung Bund in Berlin, one of Germany's largest public-sector institutions. My work involves end-to-end ownership of production ETL systems: requirements engineering, implementation, deployment, incident response, and operational handovers. The scale is real — batch loads at the billion-record level, strict enterprise change processes, and reporting systems that business departments depend on daily.
"Production-grade" in that environment means observable, restartable, and diagnosable under pressure — built for the person handling an incident at 6am, not just the person who designed it. I bring that same standard to AI: RAG, retrieval, evaluation, and agent tooling whose quality is measured and regression-gated, not asserted. And where the data demands it, the stack stays sovereign end to end — the reference architecture behind the demo on this site runs on EU infrastructure with no US service in the path.
Engineering mindset
- — I prefer long-term clarity over short-term shortcuts, especially under strict constraints.
- — Stability in production is a baseline, not a bonus. I design systems to be observable, diagnosable, and maintainable.
- — I think in systems: decompose into clear components, make the invariants explicit, and validate how the parts interact under load.
Collaboration
Working together
- — I start from a written scope: what's included, what isn't, and what success looks like.
- — Design decisions are documented as they are made. No silent architecture changes.
- — Handovers include operational runbooks, not just code.
- — I work well asynchronously. Timezone: CET/CEST (Berlin).