RAG

Retrieval-Augmented Generation: the model fetches relevant documents, then answers from them — reducing hallucination.