Retrieval-Augmented Generation
What is it?
A practice when building AI applications to help address 2 fundamental problems with LLMs:
- Spouting off answers with no source material
- Giving answers to queries that are out-of-date
It’s a system architecture pattern of introducing a separate, well-maintained data store that sits alongside your application, and injects relevant information into prompts that a given user might ask. It’s usually entirely transparent to the end user, and consists of two main parts:
- The well-structured, up-to-date data store
- The retriever that will pull relevant information from the data store and inject it into the prompt going to the LLM