RAG built for attorney–client privilege
A retrieval-augmented (RAG) system over a law firm’s case files, built to respect attorney–client privilege: strict data isolation, source traceability and no leakage to third parties.
Tech stack
- Python
- LangGraph
- pgvector
- BGE embeddings
- Claude
- FastAPI
- Zero Data Retention
- French hosting
Constraints & compliance
Map confidentiality and privilege requirements before any code. Pick French hosting and a Zero Data Retention mode.
Indexing
Semantic chunking of filings and memos, embeddings and vector storage partitioned per case. Metadata enables filtering by matter and access rights.
Sourced generation
Answers always ship with their cited passages. The model only responds from retrieved documents, never from memory.
Delivery
Production rollout, user training and guardrails: no committing action without human approval.
Screenshots
