Ground answers in evidence
RAG & Retrieval Interview Questions
Chunking, embeddings, hybrid search, reranking, citations and failure modes. How retrieval actually decides whether a RAG answer is right.
40 questions3 easy29 medium8 hard0/40 learned
01What is retrieval-augmented generation?Easyrag · retrieval · grounding02RAG vs fine-tuning: how do you choose?Mediumrag · fine-tuning · trade-offs03What are the stages of a RAG pipeline?Easyrag · architecture · pipeline04When is RAG the wrong approach?Mediumrag · design · trade-offs05RAG vs long context windows: what actually differs?Mediumrag · long context · cost06How is RAG different from building a search engine?Mediumrag · search · ranking07How do you parse documents for a RAG pipeline?Mediumparsing · ingestion · documents08What chunking strategies exist and when do they work?Mediumchunking · ingestion · retrieval09How do you pick chunk size and overlap?Mediumchunking · tuning · retrieval10What metadata should every chunk carry?Mediummetadata · permissions · citations11How do you keep a RAG index fresh?Mediumindexing · freshness · operations12How do you deduplicate documents and chunks?Mediumdeduplication · quality · retrieval13How do you handle tables, images and scanned PDFs?Mediumparsing · tables · multimodal14How do you choose an embedding model?Mediumembeddings · models · retrieval15What is the trade-off in embedding dimensions?Mediumembeddings · dimensions · performance16Cosine similarity vs dot product vs Euclidean distance?Mediumsimilarity · vectors · indexes17When do you need a domain-specific embedding model?Hardembeddings · domain · fine-tuning18What happens when documents and queries drift apart?Hardembeddings · drift · monitoring19What is multi-vector retrieval and when does it help?Hardembeddings · retrieval · late interaction20How does vector search work?Easyvector search · indexes · similarity21Why do production systems use approximate nearest neighbour indexes?Mediumann · indexes · performance22HNSW vs IVF: how do you choose an index?Hardhnsw · ivf · indexes23How do you combine BM25 and vector search?Mediumhybrid search · bm25 · fusion24How does BM25 scoring work?Mediumbm25 · lexical search · ranking25When should you rewrite the user's query?Mediumquery rewriting · retrieval · conversation26What is HyDE and when does it help?Hardhyde · query expansion · retrieval27How does metadata filtering interact with vector search?Mediummetadata · filtering · permissions28How does a reranker improve retrieval quality?Mediumreranking · precision · retrieval29What is a cross-encoder and why is it slower?Mediumcross-encoder · reranking · latency30How do you avoid returning near-duplicate results?Mediumdiversity · mmr · retrieval31How do you decide when retrieval found nothing useful?Mediumthresholds · abstention · retrieval32How do you pack retrieved chunks into the prompt?Mediumprompting · context · ranking33How do you prompt for grounded, cited answers?Mediumprompting · grounding · citations34How do you generate reliable citations?Mediumcitations · grounding · ux35Why does RAG still hallucinate?Hardhallucination · grounding · rag36How do you evaluate a RAG system?Mediumeval · retrieval metrics · faithfulness37Faithfulness vs answer relevance vs context precision?Hardeval · faithfulness · precision38What are the most common RAG failure modes?Hardfailure modes · debugging · rag39What do you monitor in a production RAG system?Mediummonitoring · operations · quality40How do you control RAG cost and latency?Mediumcost · latency · performance
Reading is step one
Now say it out loud.
Understanding an answer and delivering it under pressure are different skills. Rehearse this topic in a live RAG & Retrieval mock interview and get scored feedback on depth, clarity and confidence.