RAG (retrieval-augmented generation)

RAG (Retrieval-Augmented Generation) combines two steps: first a search component finds the most relevant passages in your documents (manuals, policies, product sheets); then the language model writes the answer using those passages as its basis. It is the usual technique behind "chatbots on company data".

Its benefits: more up-to-date answers with fewer hallucinations, and the ability to cite the source. But it shifts the problem to the data: if documents are outdated, duplicated or contain information not everyone should see, the assistant will reproduce it.

That is why a RAG project is, at heart, a data governance project: deciding which documents go in, who keeps them current and who may query what.