Case studyAI / ML
Enterprise RAG RBAC
Retrieve documents within group permissions.
- Year
- 2026
- Focus
- Permission-aware RAG
Overview
A permission-aware knowledge assistant with document ingestion, hybrid retrieval, and streamed answers with source citations.
The challenge
Document retrieval, generated answers, and citation previews must respect the requesting user's group permissions.
How it works
Parse uploaded documents and watched folders, then create structure-aware chunks for indexing.
Combine dense and keyword retrieval in PostgreSQL with ACL filtering, reciprocal rank fusion, and cross-encoder reranking.
Stream answers through FastAPI to a React interface with citation previews that recheck access permissions.
Technology stack
- Python
- FastAPI
- React
- TypeScript
- PostgreSQL
- pgvector
- RAG
Highlights
- permission-filtered retrieval
- ACL
- dense and keyword search
- Hybrid
- source passage previews
- Citations