RAG & Context Management
Retrieval-augmented generation, chunking, grounding, citations, and context-window strategy.
Before you start
- 017 multiple-choice questions, one correct answer each.
- 02Suggested time 10 minutes. The timer is a guide, not a cutoff.
- 03Use keys 1–4 to answer, arrows to move.
- 04You get a full explanation for every question at the end.
Study guide
Every question in the RAG & Context Management test, with the correct answer and a full explanation. Retrieval-augmented generation, chunking, grounding, citations, and context-window strategy. Use it to review before or after taking the timed quiz above — the answers are revealed here, so take the quiz first if you want an honest score.
Show all 7questions, answers & explanations
- RAG-01 · Question 1 of 7
What is the core idea of retrieval-augmented generation?
- AFine-tuning the model on your documents
- BRetrieving relevant context at query time and supplying it in the prompt Correct answer
- CIncreasing `max_tokens`
- DLowering temperature to 0
Why: RAG fetches relevant passages from an external store at query time and injects them into the prompt as grounding, letting the model answer from current, proprietary, or large corpora without retraining.
- RAG-02 · Question 2 of 7
Why add a short, document-aware summary to each chunk before embedding (contextual retrieval)?
- ATo reduce the embedding dimension
- BTo preserve surrounding context so isolated chunks remain meaningful and retrievable Correct answer
- CTo bypass the context window
- DTo avoid using an embedding model
Why: Chunks stripped of their document context retrieve poorly. Prepending a brief, situating summary to each chunk before embedding improves retrieval accuracy by keeping each chunk self-explanatory.
- RAG-03 · Question 3 of 7
To make answers auditable, what should a grounded RAG system ask the model to produce?
- ALonger answers
- BCitations or quotes pointing to the source passages used Correct answer
- CHigher temperature output
- DA second opinion from another model
Why: Requiring citations or supporting quotes tied to retrieved passages makes responses verifiable and discourages unsupported claims, which is essential for trust in production RAG.
- RAG-04 · Question 4 of 7
Retrieval quality is poor because chunks are too large and mix unrelated topics. What is the most direct fix?
- AIncrease `max_tokens`
- BUse smaller, semantically coherent chunks with some overlap Correct answer
- CRaise temperature
- DRemove the system prompt
Why: Oversized, topic-mixed chunks dilute embeddings. Smaller, coherent chunks (often with slight overlap to avoid cutting ideas at boundaries) embed and retrieve more precisely.
- RAG-05 · Question 5 of 7
When the relevant knowledge base fits comfortably within the context window, what is a valid simplification over a vector database?
- AFine-tuning instead
- BPlacing the full corpus directly in the prompt, ideally with prompt caching Correct answer
- CUsing a smaller model
- DDisabling retrieval entirely and relying on training data
Why: If the corpus fits the context window, you can skip retrieval infrastructure and put the whole corpus in the prompt. Prompt caching keeps this cost-effective across repeated queries.
- RAG-06 · Question 6 of 7
In a vector-search RAG pipeline, what does the embedding model actually produce?
- AA natural-language summary of each chunk
- BA numeric vector capturing semantic meaning, so similar text lands near in vector space Correct answer
- CA keyword index of exact terms
- DA compressed copy of the document
Why: Embeddings map text to dense numeric vectors whose distances reflect semantic similarity. At query time the question is embedded and nearest-neighbor search returns the most semantically relevant chunks, capturing meaning beyond exact keyword overlap.
- RAG-07 · Question 7 of 7
Pure vector search misses results that depend on exact terms like error codes or product SKUs. What is a common remedy?
- AIncrease `temperature`
- BHybrid search combining semantic (vector) retrieval with keyword/lexical search Correct answer
- CSwitch to a larger generation model
- DRemove the embedding step
Why: Hybrid retrieval runs semantic and lexical (e.g. BM25/keyword) search together and merges the results, so exact identifiers are caught by keyword matching while conceptual matches come from embeddings. A reranking step often refines the merged set.