Building a hybrid Unity + cloud AI pipeline for real-time translation, OCR-driven text-to-speech, live subtitling, and personalised AI conversations across 37 languages — inside an institution where the roadmap had to survive shifting curatorial priorities and multiple stakeholder layers.

Owned architecture and roadmap end-to-end. Designed a modular AI backend that decoupled content ingestion from the visitor layer — no manual updates, pluggable accessibility modes, clean separation of concerns. Launched publicly at the Gropius Bau in 2024. Cancelled by incoming museum direction; codebase handed over intact.