Relationship to RAG: This implements RAG's core principle of "retrieve from knowledge base rather than generate from LLM memory" - but instead of using retrieved context to generate an answer, it ...