Official German documents are dense, jargon-heavy, and often intimidating — especially for non-native speakers. The core technical problem was building a pipeline that could reliably handle low-quality camera input, extract meaningful text from varied document layouts, and return structured, actionable output in near real-time, all within a lightweight web app with no native client to rely on.

Built a full-stack web application that chains an OCR pre-processing step with an LLM-based analysis layer, returning a translation, plain-language summary, and recommended next steps in a single response. The architecture was kept deliberately lean — a server-side API handling document intake and model orchestration, paired with a minimal frontend optimised for mobile capture — to keep latency low and the experience frictionless.