Make knowledge searchable
Split source documents into useful chunks and build a search index. Embeddings can represent chunks for vector search alongside keyword retrieval.
Retrieval augmented generation
How retrieval adds relevant context to a language model’s answer.
Process overview
- Prepare
- Ask
- Retrieve
- Ground
- Answer
Steps
5 stepsThe user brings a question
The application turns the question into a retrieval query. Keyword and vector search can work together to find relevant information.
Find useful context
The retrieval system selects relevant chunks. Ranking, result limits, and permission filters help determine which context reaches the model.
Add context to the prompt
The application combines the user’s question, selected context, and instructions. This supplies information without retraining the language model.
Generate an answer with provenance
The model uses that context to produce a response. Source references help users inspect the evidence; retrieval does not guarantee a correct answer.
Keyboard: ←→
Scope
A classic RAG pattern, simplified. Search quality, access control, evaluation, and citation accuracy determine how useful the system becomes.
