Retrieval augmented generation

Retrieval augmented generation, or RAG, lets a language model answer using your organisation's own knowledge. Relevant passages are retrieved from your documents or databases and given to the model, so answers are grounded, current and can cite their sources.

1. 1

   Index

   Documents are split into passages and stored with embeddings in a vector or hybrid search index.

2. 2

   Retrieve

   For each question, the most relevant passages are found using semantic and keyword search.

3. 3

   Generate with citations

   The model answers from those passages and links back to the sources.

• Internal knowledge assistants for staff
• Customer help centres that answer in plain language
• Searching contracts, policies and manuals
• Support agents grounded in product documentation

Good retrieval matters more than the model. We focus on clean source data, hybrid search, access controls that respect permissions, and evaluation sets that measure answer quality.