Retrieval-Augmented Generation

Also called RAG

A way of getting a language model to answer from your own information: the system first searches your documents for the relevant passages, then gives them to the model along with the question so its answer is based on them.

Why it matters

It lets a product answer from current, private or company-specific knowledge without retraining a model, and lets the answer point to its sources. Most of the quality comes from the search step, so poor retrieval produces confident but wrong answers.

For example

An HR startup's assistant answers employees' questions by first pulling the relevant section of the company's leave policy, then quoting it in the reply.

Go deeper

What is RAG (Retrieval-Augmented Generation)? AWS · article AWS's plain explainer of how retrieval-augmented generation works, why it is used instead of retraining a model, and how it lets answers cite their sources. Open aws.amazon.com →

Related terms

Also in Starting Up

See how founders actually handle this on Building the product, part of the Starting Up hub.

eChai Partner Brands The partnership →