RAG Interview: Our Search Returned the Wrong Document. Why?

Twelve RAG interview questions with a small runnable lab: chunking, hybrid search and rank fusion, retrieval metrics (recall at k, MRR), why RAG still hallucinates, re-indexing, per-document access control, agentic RAG, long context versus RAG, and the cost budget. The retrievers in the lab are word-based stand-ins, and you'll see exactly where that fails. Part 7 of AI Engineer Interview Kit 2026. Chapters 0:00 Intro 0:46 1. RAG end to end, and where each stage fails 1:20 2. Choose a chunking strategy: how would you know it's wrong? 2:27 3. Choose an embedding model, and survive changing it 3:10 4. Hybrid search and reranking: when does fusion not help? 4:11 5. pgvector or a dedicated vector DB: what decides it? 5:18 6. Retrieval looks right, the answer is wrong. Why? 6:04 7. Evaluate retrieval separately from generation 6:51 8. A document is edited or deleted. What does the index do? 7:34 9. Enforce per-document access control in RAG 8:09 10. When does agentic RAG beat a fixed pipeline? 8:46 11. Million-token windows exist. Why keep retrieval? 9:28 12. Budget a RAG query: where do money and time go? Code for this video: https://github.com/codeandagents/inte... Versions used: Python 3.14 (standard library only) Sources: https://plg.uwaterloo.ca/~gvcormac/co... https://arxiv.org/abs/2307.03172 https://www.anthropic.com/news/contex... https://github.com/pgvector/pgvector https://docs.ragas.io/ About the channel: tutorials by a software architect with 20+ years of designing and building highly scalable systems. Narration uses an AI voice; the code, the runs on screen and the explanations are the author's own.