A RAG system that demystifies German bureaucratic correspondence (Behördenbriefe). Users photograph an official letter; OCR extracts the German text, pgvector semantic search retrieves relevant context from a curated knowledge base, and an LLM explains what the letter means, what action is required, and by when — in the user's preferred language. Demonstrates a grounded RAG pipeline reducing hallucination on legally consequential content. Stack: Next.js, FastAPI, OCR, GPT-4o-mini, pgvector RAG, Dodo Payments. URL: https://behoerdenbot.sbs