Concept AI Concepts

What Is RAG? Letting AI Take the Open-Book Exam

A plain-language visual explainer on retrieval augmented generation: why models cannot know your documents, how the right pages get found and handed over, and why the answer is only as good as the library.

Intermediate

3 min

9 slides

9 views

August 4, 2026

Mstimaj stands between a glowing open book and a locked cabinet, presenting the open one.

A model sitting on its own is taking a closed-book exam. It answers from memory, and when memory runs out it guesses fluently. RAG is the change that hands it the book first. It is the reason an assistant can answer questions about your own documents, your own policies, and things that happened after it finished training.

AI Concepts

What Is RAG? Letting AI Take the Open-Book Exam

What is RAG?

Retrieval augmented generation is how an AI answers from your documents instead of guessing from memory. Seven steps, in plain language.

Technical: RAG stands for retrieval augmented generation.

Metaphor: An open book and a locked cabinet side by side, the book lit and waiting

On its own, it is a closed-book exam

A model answers from what it absorbed during training and nothing else. No notes, no lookups. When the question lands outside what it studied, it does not go quiet. It writes something fluent and plausible anyway.

Technical: Training data has a cutoff date, and nothing after it exists to the model.

Metaphor: A figure at an exam desk with the books locked away behind glass

It was never going to know your things

Your contracts, your client notes, your internal policies, last Tuesday's meeting. None of that was in the training data and none of it ever will be. This is not a gap the next bigger model closes.

Technical: Private and recent information is absent by definition, regardless of model size.

Metaphor: A locked private cabinet standing apart from the public shelves

RAG hands it the book

Instead of asking the model to remember, you let it look things up. Find the relevant pages first, put them in front of the model, then ask the question. Closed book becomes open book.

Technical: RAG stands for retrieval augmented generation.

Metaphor: The glass lifting and an open book sliding onto the exam desk

Your documents get filed by meaning

First everything is broken into passages and filed by what it is about, not by keyword. That way a question about time off still finds the page headed holiday policy, even though the words do not match.

Technical: Passages are stored as embeddings, which place similar meanings near each other.

Metaphor: Pages sorting themselves into a glowing map where related ones cluster

Your question goes and fetches

You ask something. Before the model writes a word, the system searches that filed library and pulls out the handful of passages closest to what you asked. Not the whole library. Just the pages that matter.

Technical: Retrieval typically returns a small ranked set of the closest passages.

Metaphor: A beam sweeping a library and lifting out a few glowing pages

Now it answers with the pages open

Those passages go to the model along with your question. It writes the answer from what is in front of it rather than from memory, which is why a good setup can point at the exact document it used.

Technical: Grounding an answer in retrieved text is what makes citation possible.

Metaphor: A machine composing an answer with open pages held beside it

A messy library gives messy answers

RAG does not make a model truthful. It makes it accurate about whatever you handed it. Outdated policies, duplicate files, and a three-year-old price list will be quoted back to you with total confidence. The library is the product.

Technical: Retrieval quality sets the ceiling on answer quality.

Metaphor: A neatly kept shelf beside a disordered one, both feeding the same machine

RAG in five lines

The whole idea in five lines: a model only knows its training, your files were never in it, so you file them by meaning, fetch the closest passages, and accept that answer quality follows library quality.

Technical: Retrieval quality sets the ceiling on answer quality.

Metaphor: A tidy glowing shelf with five labelled bands of light running across it

1 / 9

Want to go deeper?

Let's Work Together

Whether you need AI automation, strategic guidance, or want to explore what's possible, I'm here to help.

Get in Touch Learn About Me

What Is RAG? Letting AI Take the Open-Book Exam

What Is RAG? Letting AI Take the Open-Book Exam

What is RAG?

On its own, it is a closed-book exam

It was never going to know your things

RAG hands it the book

Your documents get filed by meaning

Your question goes and fetches

Now it answers with the pages open

A messy library gives messy answers

RAG in five lines

More from Neural Canvas

What Is a Context Window? Why Long Chats Start Forgetting

What Is an AI Agent? The Difference Between Answering and Doing

What Is MCP? The Universal Adapter for AI Tools

Let's Work Together