RAG teaches the material. The Q&A Quick Reference drills the core 28 questions most interviews actually ask. This page goes further — a deep, comprehensive bank covering the long tail of RAG research architectures, production deployment and cost, monitoring and incident debugging, security and multi-tenancy, and open-ended system-design scenarios, pulled from a dedicated 1,500+ question interview-prep repository and cross-checked against other recent interview-question roundups. The goal is coverage wide and deep enough that no reasonable RAG question — fundamentals, an obscure architecture name, or "walk me through debugging this production incident" — should catch you with nothing to say.
This page assumes you already know everything on the /docs/llms-genai/rag page and the core 28 questions — it doesn't re-teach chunking or BM25 from scratch, it builds on top and goes sideways into territory those two pages don't cover.
You will not get asked all 117 of these in one interview. The point isn't memorizing every answer verbatim — it's recognizing the shape of question each category represents, so an unfamiliar specific question in a familiar category still gets a structured, confident answer instead of a blank.
Questions with a checkable answer (a diagnosis, a decision, a number) check what you pick or type. The rest are explanations: answer out loud, then open the card.
26 checkable questions on this page. Each is checked when you answer, and the worked answer opens after.
RAG Fundamentals & Strategy
A RAG answer is wrong even though the correct document exists somewhere in the corpus. How do you debug it?
A 300-page policy manual is updated monthly. Answers must cite the exact clause, and the bot should write in your brand's voice. Which combination fits?
Chunking & Preprocessing (Advanced)
Fixed 500-character chunks over a Python codebase keep cutting functions in half. What should the chunking unit be?
A chunk exceeds the embedding model's max input length. What actually happens, and why is this a dangerous failure mode?
Embeddings & Vector Search (Advanced)
How much memory do 1 billion 768-dimensional float32 vectors take, uncompressed? (TB, where 1 TB is 10¹² bytes.)
Your index will hold 1.5 billion vectors. You have one node with 64 GB of RAM and a fast SSD, and 5 ms latency is acceptable. HNSW or DiskANN?
Given the choice, should you fine-tune the reranker or the embedding model first?
Why do hard negatives matter more than random negatives when fine-tuning an embedding model with contrastive loss?
Retrieval Architectures — The Research Catalogue
This is the single richest category — a question on every named architecture a RAG-focused interview might drop on you, beyond the ones already covered on the main RAG page (Agentic RAG, CRAG, Self-RAG, FLARE, RAPTOR, CAG, Multimodal RAG, SQL RAG, Code RAG, GraphRAG). You don't need to have implemented all of these — you need to recognize the name and state the one mechanism that actually distinguishes it.
Each hop in a 4-hop retrieval chain is 90% reliable, independently. Roughly how reliable is the whole chain? (As a decimal.)
A RAG answer has a citation next to every sentence. Which check tells you the claims are actually supported?
Why does standard RAG underperform specifically on tables, and what does Table-Aware RAG do differently?
Agentic & Self-Correcting RAG (Beyond the Basics)
How do you prevent an agentic RAG loop from running forever (or effectively forever, on your bill)?
GraphRAG & Structured Data
Which query needs a knowledge graph rather than a vector index?
Which question needs GraphRAG's global search rather than local search?
Conversational & Memory RAG
Turn 1: "Tell me about the March outage." Turn 3: "Did it affect enterprise customers?" Retrieval for turn 3 returns unrelated pages. What is the most likely cause?
Multimodal & Specialized RAG
Evaluation & Testing (Advanced)
Your thumbs-up rate is 92%. Does that mean your RAG system is actually good?
Your offline eval scores improved after a change, but users are complaining more. What's going on?
Your LLM-judge scores dropped 5% overnight with no deploy on your side. What do you check first?
RAG in Production — Deployment & Scaling
Where does the money actually go in a RAG system's per-query cost, and which stage dominates?
Each request has a 6,000-token fixed prefix and 500 tokens that vary. Cached input tokens cost 10% of the normal price. If every request hits the cache, by what percentage does the input cost fall?
What's a cache stampede, and how do you prevent one in a RAG caching layer?
Which of these should you not cache?
RAG in Production — Monitoring, Debugging & Incidents
Your RAG system's p99 latency just doubled with no obvious deploy. How do you debug it?
RAG in Production — Security & Compliance
Retrieval is correctly filtered by permissions. Then a "related documents" feature follows links from the retrieved chunks to other documents. What's the risk?
An approximate search returns the top 100 candidates, then a permission filter keeps only documents this user may see: 2% of the corpus. If permissions are unrelated to relevance, how many results are left on average?
System Design / Scenario Questions
How would you migrate a production RAG system to a new embedding model with zero user-facing downtime?
Next: Q&A Quick Reference for the shorter breadth-round drill, or back to RAG for the full teaching page any of these answers builds on.