Free guide

RAG: a practical guide.How AI answers from your documents, and how to check it’s right.

Written for owners and managers deciding whether an assistant that answers from company documents is worth building. Plain English, no hype, and honest about where it goes wrong.

The architecture. What happens before anyone asks a question, and what happens every time someone does.
A worked example. One question followed end to end, then seven ways the same system goes wrong.
A testing checklist. How to test an assistant before launch and after every change, with a question-set template.

Free PDF · 60 pages

RAG: a practical guide

+ question-set template

We’ll email you the guide and record your request. Marketing emails only if you tick the box, and you can unsubscribe at any time. See our privacy notice.

The short version

What is RAG?

RAG (retrieval-augmented generation) is a way of making an AI assistant answer from your own documents instead of from memory. When someone asks a question, the system first finds the relevant passages in your approved material, then asks the AI to write an answer using only those passages, and shows where each part came from. If the right passage isn’t there, a well-built system says so rather than guessing.

How does an assistant answer from my documents?

Your documents are split into passages and indexed ahead of time. At question time the system retrieves the most relevant passages, checks the asker is allowed to see them, and drafts an answer with links back to the source.

When is RAG not the answer?

When fixed rules would do the job, when the information isn’t written down anywhere, or when an existing product already answers the question well. I’ll tell you if a simpler tool is the better start.

How do I test one before relying on it?

Build a set of real questions with the expected source for each, run the system against them, and include questions it should decline. The guide includes the question-set template and a checklist.

What does it need from my data?

Documents that are current, owned by someone and stored where access can be controlled. Messy files are workable; contradictory or out-of-date ones are the real problem.

Knowledge assistants covers how I build and hand over these systems, and a free first call is the quickest way to find out whether one fits.