DocAssist RAG over real-estate documents

Retrieval-augmented assistant for property audits, energy certificates and contracts: PDF/DOCX/XLSX ingestion, Qdrant vector index, Azure OpenAI answers with source attribution and confidence scores.

DocAssist answers questions over the documents real-estate professionals actually handle: building audits, energy assessments, construction specifications and contracts. Files in PDF, DOCX, TXT and XLSX are chunked, embedded and stored in Qdrant; a FastAPI service retrieves the relevant passages and Azure OpenAI writes the answer.

Every answer carries the source documents it came from and a confidence score derived from retrieval quality, so a user can check the claim instead of trusting it. Indexing is incremental (only new or modified files are processed) and domain categories (energy, water, transport, regulatory) shape the retrieval.

The point of the project was production discipline around RAG rather than a demo: Docker packaging, CI checks, logging, and honest fallbacks when retrieval is weak. Source code is public on GitHub.