Retrieval-Augmented Generation

Retrieval-augmented generation combines a search step with a generation step. When a question is asked, the system retrieves the most relevant passages from a defined document set, then instructs the language model to answer using only that retrieved material.

The approach reduces fabrication because the model works from supplied source text rather than parametric memory, and it allows every statement in an answer to be traced back to a specific document and page.

Alternative Names:

RAG, Grounded Generation

Why it Matters?

In litigation, the value of an AI answer depends entirely on whether it can be verified against the record. RAG architectures make citation to the underlying medical record, deposition, or production document possible, which is what allows attorney review to be efficient rather than a full re-reading of the source. Retrieval quality, not model size, is usually the limiting factor on accuracy in matter-specific work.

Frequently Confused with

Related terms

Frequently asked questions

Does retrieval-augmented generation eliminate hallucinations?

No. It substantially reduces them by grounding answers in retrieved text, but errors can still occur when retrieval misses relevant passages or the model overstates what a source says. Verification remains necessary.

How is RAG different from fine-tuning?

Fine-tuning changes the model's weights to adjust its behavior. RAG leaves the model unchanged and instead supplies relevant documents at question time, which is better suited to matter-specific facts that change constantly.