Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation na technique wey ground language model output in retrieved source documents instead of relying only on im parametric memory. Di system fetch relevant passages from knowledge base, supply dem as context, and ask model to answer using only dat evidence.
Synonym dem: RAG, retrieval augmented generation, grounded generation, context augmentation
Retrieval-augmented generation separate wetin model know from wetin dem train am on. At query time retriever pull most relevant chunks from maintained knowledge base, and model compose answer constrained to dat evidence. This reduce hallucination, make answers auditable through citations, and let teams update knowledge without retraining. For governed setting, RAG na mechanism wey turn question into grounded, cited response wey person or policy fit verify before trust.