← Back to Projects

UK Company Filings RAG: Grounded Financial Q&A System

A retrieval-augmented generation system for querying UK annual reports and filings from Companies House, built to solve RAG’s biggest failure mode in finance: hallucinated numbers. Retrieved financial claims are verified against source tables before being surfaced, with every answer citing the exact filing, section, and table cell it came from.

A well-tested prototype rather than a production deployment: an automated evaluation harness (89% pass rate) gates every CI run, backed by 119 unit tests and structured JSON request logging.

Python, FastAPI, hybrid retrieval (vector search + BM25), Chroma, Docker, GitHub Actions (CI), Prometheus, Companies House API