13.3 Capstone 3 — RAG over a real corpus
A grounded question-answering system with citations and an eval suite.
The third capstone is a grounded question-answering system over a real corpus: retrieval plus reranking with measured quality, citations enforced, refusal when no evidence exists, and a golden-set regression suite run on every change. The eval suite is the differentiator, because most candidates cannot show one. The design test is refusal — a system that answers confidently when retrieval found nothing has failed at exactly the moment trust was the product.
Work through these
Corpus selection, ingestion, chunking
Choosing a document collection, getting it in, and deciding how to cut it up. The chunking decision quietly determines the ceiling on everything that follows.
Retrieval plus reranking; measured retrieval quality
Retrieving candidates and reordering them, with the retrieval quality measured on its own. Measuring retrieval separately is what makes the system debuggable.
Citation enforcement and refusal on no-evidence
Requiring answers to cite their sources, and refusing when nothing relevant was found. The refusal behaviour is what makes the system trustworthy.
Golden set with regression runs on every change
A fixed set of questions with known good answers, run on every change. This is what turns improvements from impressions into evidence.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.