Retrieval-Augmented Generation (RAG)

Letting the AI look things up in your documents before it answers.

In everyday terms

Instead of relying on memory, the system first searches a knowledge base, pastes the relevant bits into the prompt, then the AI answers using them, ideally with sources.

For professionals

Retrieve top-k chunks (often via embeddings), inject into context, generate a grounded answer. Quality hinges on chunking, retrieval and citation.

Think of it like…

An open-book exam instead of a closed-book one.

You've already seen it

Company chatbots that answer from internal policies, AI search with citations.

Myth vs reality

Myth: RAG means the AI is retrained on your documents.

Reality: Nothing is retrained. Relevant text is fetched and shown to the model at question time.

Quick check

What does RAG add before the AI answers?

A retraining step
A search for relevant information
A higher temperature
A human reviewer

Show answer

A search for relevant information: Retrieve, then generate.

Builds on

Embedding · Context Window

Embedding · Vector Database · Hallucination · Context Window