Skip to main content

Retrieval-Augmented Generation (RAG)

RAG is a technique that grounds AI outputs in specific, external data rather than relying solely on the model's internal knowledge.

The Problem​

Models have a "cutoff date" for their knowledge and can hallucinate when they lack information.

The RAG Solution​

  1. Retrieve: Search a knowledge base for relevant documents.
  2. Augment: Inject those documents into the prompt.
  3. Generate: The model answers based only on the provided context.

Standard RAG Prompt Pattern:​

Answer the question based ONLY on the provided context. If the answer isn't there, say "I don't know."

Context: [Retrieved Text Here]

Question: [User Question Here]