RAG (Retrieval-Augmented Generation) Practice Test

RAG (Retrieval-Augmented Generation) Practice Test: all 15 questions with answers

Try each question first, then open “Show answer” to check the correct option and explanation.

1. Q1.What does retrieval-augmented generation (RAG) do?

   • A.Retrains the model every day
   • B.Retrieves relevant documents and adds them to the prompt so the model answers from them
   • C.Generates random documents
   • D.Compresses the model weights
   Show answer

   Correct answer: B. Retrieves relevant documents and adds them to the prompt so the model answers from them

   RAG combines search with generation so answers are grounded in your own data.

2. Q2.What is the main benefit of RAG over relying on the model alone?

   • A.Answers can use private or up-to-date information without retraining the model
   • B.It removes the need for any data
   • C.It makes the model smaller
   • D.It guarantees zero errors
   Show answer

   Correct answer: A. Answers can use private or up-to-date information without retraining the model

   You update the document store, not the model, when knowledge changes.

3. Q3.What is chunking?

   • A.Deleting old documents
   • B.Splitting documents into smaller pieces for indexing and retrieval
   • C.Combining many models
   • D.Encrypting data
   Show answer

   Correct answer: B. Splitting documents into smaller pieces for indexing and retrieval

   Retrieval works on chunks, so chunk size and boundaries strongly affect answer quality.

4. Q4.What is a common risk of making chunks very small?

   • A.They lose the surrounding context needed to answer the question
   • B.They use too much storage
   • C.They cannot be embedded
   • D.They make search impossible
   Show answer

   Correct answer: A. They lose the surrounding context needed to answer the question

   Tiny chunks often split a fact from the context that explains it. Very large chunks have the opposite problem: less precise matches.

5. Q5.Why do many pipelines add overlap between chunks?

   • A.To double the cost
   • B.To keep information that spans a chunk boundary together in at least one chunk
   • C.To remove duplicates
   • D.Overlap is never useful
   Show answer

   Correct answer: B. To keep information that spans a chunk boundary together in at least one chunk

   A small overlap stops sentences or ideas from being cut in half at chunk edges.

6. Q6.What does a vector database mainly store and search?

   • A.Embeddings, so similar items can be found quickly
   • B.Only images
   • C.User passwords
   • D.Spreadsheet formulas
   Show answer

   Correct answer: A. Embeddings, so similar items can be found quickly

   Vector databases index embeddings for fast nearest-neighbour similarity search.

7. Q7.What is the usual order of a basic RAG query?

   • A.Generate the answer, then search
   • B.Embed the question, retrieve similar chunks, add them to the prompt, then generate
   • C.Fine-tune, then search
   • D.Search the web, then delete results
   Show answer

   Correct answer: B. Embed the question, retrieve similar chunks, add them to the prompt, then generate

   Retrieval happens before generation so the model can use the retrieved context.

8. Q8.Which embedding model should be used for queries?

   • A.A different one from the documents, to add variety
   • B.The same embedding model used to embed the documents
   • C.No model; use raw text
   • D.Any image model
   Show answer

   Correct answer: B. The same embedding model used to embed the documents

   Vectors from different embedding models live in different spaces and cannot be compared meaningfully.

9. Q9.What is hybrid search?

   • A.Searching two databases at random
   • B.Combining keyword search (such as BM25) with semantic vector search
   • C.Searching only by date
   • D.Using two LLMs
   Show answer

   Correct answer: B. Combining keyword search (such as BM25) with semantic vector search

   Keyword search catches exact terms like product codes; vector search catches meaning. Together they improve recall.

10. Q10.What does a reranker do?

    • A.Retrains the embedding model
    • B.Re-scores the retrieved candidates with a more precise model so the best ones come first
    • C.Deletes irrelevant documents permanently
    • D.Translates the query
    Show answer

    Correct answer: B. Re-scores the retrieved candidates with a more precise model so the best ones come first

    Rerankers (often cross-encoders) are slower but more accurate, so they are applied to a short candidate list.

11. Q11.Why ask a RAG system to cite its sources?

    • A.Citations make the answer longer
    • B.Users can check each claim against the original document
    • C.It lowers the token cost
    • D.It is required by law everywhere
    Show answer

    Correct answer: B. Users can check each claim against the original document

    Citations build trust and make wrong answers easier to spot.

12. Q12.A RAG bot gives wrong answers. Logs show the retrieved chunks are irrelevant. What should you fix first?

    • A.The model's temperature
    • B.Retrieval: chunking, embeddings, filters or search settings
    • C.The colour of the chat UI
    • D.Nothing; this is normal
    Show answer

    Correct answer: B. Retrieval: chunking, embeddings, filters or search settings

    If the right context is never retrieved, no prompt can make the model answer correctly.

13. Q13.What is metadata filtering in retrieval?

    • A.Removing all metadata
    • B.Restricting search by attributes such as date, department or document type
    • C.Sorting results alphabetically
    • D.Encrypting metadata
    Show answer

    Correct answer: B. Restricting search by attributes such as date, department or document type

    Filters narrow the search space, for example to only this year's HR policies.

14. Q14.What does recall@k measure for retrieval?

    • A.How often a relevant chunk appears in the top k results
    • B.How many tokens the answer uses
    • C.How fast the model responds
    • D.The number of documents stored
    Show answer

    Correct answer: A. How often a relevant chunk appears in the top k results

    Recall@k checks whether retrieval surfaces the right evidence before generation even starts.

15. Q15.When is fine-tuning a better choice than RAG?

    • A.When knowledge changes every day
    • B.When you mainly need to change style, format or behaviour rather than add changing facts
    • C.When you need citations
    • D.Never
    Show answer

    Correct answer: B. When you mainly need to change style, format or behaviour rather than add changing facts

    RAG is best for knowledge that changes; fine-tuning is best for consistent behaviour or format.