A technique where an AI system looks up relevant information from a specific set of documents before generating a response, so its answers are grounded in real, current information.
Definition of “RAG (Retrieval-Augmented Generation)”
Language models are trained on general data and do not automatically know a specific business's policies, pricing, or inventory. RAG solves this by having the AI system search a connected set of documents, such as a knowledge base or product catalog, for relevant information before writing its response.
The result is an answer that combines the model's language ability with facts pulled from the business's own, current information, rather than relying on what the model happened to learn during training, which may be outdated or simply wrong for that business.
“RAG (Retrieval-Augmented Generation)” In Practice
A company's support assistant uses RAG to search the current product documentation before answering a question about a specific feature, ensuring the answer reflects the latest version of the product rather than an older one the model might have seen during training.