Retrieval-augmented generation

At a glance

Abbreviation

RAG

Core stages

Retrieve source material, then generate

Overview

Retrieval-augmented generation combines an information-retrieval step with a generative model. The system finds potentially relevant source material and supplies it as context for producing an answer, so the response can draw on information outside the model's training data.

The retrieval boundary

A retrieval system searches an indexed collection and returns candidate passages or records. Their identifiers and provenance should travel with the content so that an answer can point back to the material it used.

The answer boundary

Supplying sources does not guarantee a correct answer. Evaluate whether retrieval found suitable evidence and whether the generated response represents that evidence accurately, including when the available material cannot answer the question.

Sources and review

MOOR's explanatory text is supported by the following source links.

1. Retrieval-augmented generation — Cohere
2. Retrieval guide — OpenAI