RAG

RAG (Retrieval-Augmented Generation) connects LLM reasoning with the breadth of an external knowledge base: a retriever finds relevant chunks, a generator receives them as context, and responds.

Enables use of knowledge post-training cutoff date and proprietary domain knowledge – without retraining.

Pipeline:

document chunking → embedding/indexing → retrieval → context augmentation → generation

Retriever quality defines the ceiling: if chunks aren’t found, even the best LLM is powerless.

Related: Document Chunking, Hybrid Search, Agentic RAG