What does RAG stand for?

Retrieval-augmented generation: the system retrieves relevant passages from your own content and gives them to an AI model so its answer is grounded in your data.

How much does it cost to build a RAG system?

A production RAG copilot with Softgen starts from £18,000, including ingestion, retrieval, evaluations and deployment into your product.

What Is RAG? Retrieval-Augmented Generation Guide

Key takeaways

RAG retrieves relevant passages from your own content and gives them to the model, so answers are grounded in your data.
It's the right default when knowledge changes often or answers must cite sources.
Fine-tuning changes how a model behaves; RAG changes what it knows at answer time.
Retrieval quality — chunking, search and ranking — decides most of the result.
Production RAG needs evaluations, citations, access control and monitoring.

The short answer

Retrieval-augmented generation (RAG) is a way of making an AI model answer from your information. Instead of relying only on what the model learned in training, the system first retrieves the most relevant passages from your documents, database or help centre, then gives them to the model to generate an answer — ideally with citations.

It's how a support copilot answers from your product docs, or an internal assistant answers from your policies, without making things up.

How RAG works, step by step

Ingest. Your content — documents, web pages, tickets, PDFs — is split into sensible chunks.
Index. Each chunk is turned into an embedding (a numerical fingerprint of its meaning) and stored, often alongside a keyword index.
Retrieve. When someone asks a question, the system finds the chunks most relevant to it — usually by combining semantic and keyword search, then re-ranking.
Generate. The model receives the question plus the retrieved chunks and writes an answer grounded in them.
Cite. A good system shows which sources it used, so people can check.

RAG vs fine-tuning

In one line: RAG changes what the model knows at answer time; fine-tuning changes how it behaves. Most products need RAG first — the full comparison is in RAG vs fine-tuning.

Why RAG demos work and RAG products fail

A demo on ten tidy documents looks brilliant. Production breaks on:

Bad chunking — answers split across chunks, tables mangled, context lost.
Weak retrieval — the right passage exists but isn't found.
No evaluation — nobody knows whether a change made answers better or worse.
Permissions — users seeing content they shouldn't.
Stale content — the index drifts out of date.

What production RAG needs

Hybrid search and re-ranking so the right passages come back.
Evaluations: a test set of real questions scored before every change (AI evals explained).
Citations in every answer, and a graceful "I don't know".
Access control that respects who can see what.
Monitoring and feedback so quality improves over time.

How we build RAG at Softgen

We build RAG copilots and assistants into production — with evaluations, guardrails, citations and observability — as part of our AI development work. AI builds start from £18,000 on a fixed price and date.

Related guides.

What is an AI copilot — and should you build one into your product?

Every product is bolting on a copilot. Here's what an AI copilot actually is, how it differs from a chatbot and an agent, and whether yours needs one.

How to add LLM features to your existing app

Your app already works. Adding AI that makes it noticeably better is the hard part. Here's the exact playbook we use to wire in agents, copilots and RAG without turning it into a support nightmare.

All insights

What is RAG? Retrieval-augmented generation in plain English