Multi-Collection RAG — Retrieval Across Multiple Knowledge Bases

I will build a multi-collection RAG system — routing queries to the appropriate knowledge base collection (or multiple collections simultaneously), retrieving and reranking across collections, resolving conflicts when collections provide contradictory information, and synthesising a unified response from multi-source retrieved content. Enterprise RAG systems serving multiple departments need to retrieve from domain-appropriate collections — a query about HR policy should retrieve from the HR knowledge base, a technical query from the engineering knowledge base, and some queries from multiple collections simultaneously; a routing and federation layer is what makes this possible without exposing all collections to every query.

The system covers collection schema design, query-to-collection routing classifier, parallel multi-collection retrieval where appropriate, cross-collection result fusion with conflict detection, unified synthesis prompt handling multi-source content, and quality evaluation across routing scenarios.