Open-source RAG interview guide covers designs and common failures

An open-source GitHub repository collects RAG interview questions and answers based on production experience. It covers more than 200 questions across 12 RAG designs, including naive RAG, agentic RAG, graph RAG, self-checking RAG, speculative RAG, and multimodal RAG. The questions are grouped by difficulty: 13 basic, 58 intermediate, and 129 advanced.

It also explains 6 common production failure modes, including wrong answers even with context, failed retrieval, embedding mismatch, stale indexes, context window overflow, and rerankers pushing the right answer down. The repository includes concept notes on chunking, embeddings, vector databases, reranking, evaluation metrics, and prompt injection. It also has a one-table cheatsheet comparing all 12 RAG types, plus study paths for one-week preparation, phone screens, and system design interviews.

Planned additions include labs, an interview simulator, and evaluation tools.