AI
RAG (retrieval-augmented generation)
Also called
- Retrieval augmented generation
- Grounding
A language model knows what was in its training data. It does not know your pricing, your policies or last week's incident report. RAG closes that gap: when a question arrives, the system searches your own content, selects the passages most likely to be relevant, and hands them to the model along with the question.
The retrieval half is where these systems succeed or fail, and it is the half most demos skip. If the search returns the wrong three paragraphs, a better model will produce a more fluent wrong answer. Chunking strategy, embedding choice, hybrid keyword-plus-vector search and re-ranking are all retrieval problems, not model problems.
Done properly, RAG also gives you citations. The system can show which document each claim came from, which is what turns an AI answer from a liability into something a support agent or a regulator can check.
RAG is what makes an AI assistant cite your policies instead of inventing them — the difference between a tool your team trusts and one they quietly stop using.
Commonly misunderstood
What people get wrong
The claim
“We need to fine-tune a model on our data.”
What is actually true
Almost never, and almost never first. Fine-tuning teaches style and format; retrieval supplies facts. Most projects that reached for fine-tuning needed better retrieval, and found out expensively.
Related terms
Next step
Working through a rag (retrieval-augmented generation) decision?
Tell us the situation. We will give you the tradeoffs as we see them, including when the answer is that you do not need what you are being sold.
No pitch deck. A 30-minute conversation about what you are trying to achieve.