S3E2 RAG

What this lesson is about

RAG stands for Retrieval-Augmented Generation, and it is how an AI answers questions from documents it was never trained on. You ask something, it searches the files it has been given, pulls out the most relevant passage, and writes its answer using that passage alongside what it already knew. Today you take that process apart from the inside, because the Gem you are about to open is a RAG system and it knows it. The demonstration happens to you, live, rather than being described to you.

What you'll be able to do

• I can explain what RAG is, in my own words, to someone who has never heard the term.
• I can tell the difference between a knowledge source, which gets searched when needed, and a system prompt, which is always switched on.
• I can design a question that proves whether retrieval actually happened.
• I can describe at least one way retrieval fails, and say what that failure looks like from the outside.

A little background

The two halves of the name do the work. Retrieval is the search step: your files are cut into passages called chunks, and the ones closest in meaning to your question get pulled out. Generation is what happens next, when the model writes an answer with those chunks in front of it. Everything interesting about RAG lives in the seam between the two, because a model that retrieves the wrong chunk will still write a confident answer from it.

The Gem's corpus is invented on purpose. Nothing in it exists anywhere else, so if it can quote the document back to you, retrieval demonstrably happened — there is nowhere else the answer could have come from. That is the check that confirms the lesson landed, and it is the same check you can run on any RAG system you meet later.