Designing Governed RAG on Data Products

This article details a governed RAG architecture that separates retrieval and governance to build trustworthy enterprise AI assistants. It covers a three-layer design: a retrieval layer using vector stores from governed data products, a structured query layer routing through semantic layers for deterministic SQL, and a governance layer enforcing access policies. Key topics include data products as retrieval sources, access policy enforcement, semantic layer integration, audit trails, and chunking strategies. The approach ensures answers are accurate, fresh, and user-appropriate, avoiding common pitfalls like stale data or unauthorized access.