Curriculum
Retrieval
Retrieval is how most AI products turn private documents into grounded answers, but PMs need to know where it fails and how to prove it works. You build a RAG pipeline end to end, score retrieval apart from the answer, compare keyword, vector, hybrid, and reranked search on labeled questions, and add permission-aware filtering with a leak test.
- 09.01
Build a RAG pipeline end to end
You build every stage of a retrieval pipeline over your own documents, break one stage on purpose, and decide when retrieval beats putting everything in a long context window. - 09.02
Measure retrieval, then pick the search
You measure retrieval quality separately from answer quality, then compare keyword, vector, hybrid and reranked search on your own labeled questions and pick one on evidence. - 09.03
Permission-aware retrieval
Design and test a retrieval layer for your Quillnote help chatbot that filters chunks by the asker's permissions before the model sees them, and price the stale-access window a permission change creates.