Prepstellar
Get it on Google Play

AI-103 · Practice set 8 of 13

Retrieval and Indexing Methods: 10 practice questions

10 questions · Untimed · Free

10 free AI-103 practice questions on Retrieval and Indexing Methods, with an explanation for every answer. Untimed. The full mock exam and the timed version are in the app.

Set 8 · Retrieval and Indexing Methods · 10 questions Read the lesson

  1. Question 1 of 10

    What is the order of the three steps in a RAG request?

    1. ARetrieve, augment, generate
    2. BAugment, retrieve, generate
    3. CGenerate, retrieve, augment
    4. DRetrieve, generate, augment
    Show the answer

    The order follows the data. First the app finds relevant content, then it puts that content into the prompt next to the question, and only then does the model answer. Generating before retrieving would mean answering without the grounding data.

    Next → 1 / 10

  2. Question 2 of 10

    Which Azure AI Search query form combines text search and vector search in a single request?

    1. AFull-text search
    2. BFuzzy search
    3. CHybrid search
    4. DVector search
    Show the answer

    Hybrid search runs keyword and vector matching together in one request. Text search works on plain-text searchable fields and vector search on vector fields, so you get exact-term matches and meaning-based matches in one result set.

    Next → 2 / 10

  3. Question 3 of 10

    A support bot must answer from a product catalog that changes every week. Which pattern fits?

    1. AA longer system message with prices
    2. BA larger model with more parameters
    3. CFine-tuning on each weekly catalog
    4. DRAG over an index of the catalog
    Show the answer

    Fresh knowledge belongs in an index, not in model weights. RAG grounds answers in private or frequently changing data, so updating the index updates the answers. Fine-tuning is for changing behavior, style, or task performance.

    Next → 3 / 10

  4. Question 4 of 10

    Which features of Azure AI Search can a RAG pipeline usually skip because they serve people typing in a search box?

    1. AAutocomplete and suggestions
    2. BHybrid queries and scoring profiles
    3. CChunking and integrated vectorization
    4. DSemantic ranking and synonym maps
    Show the answer

    Some search features exist for human interaction and add nothing to grounding. Autocomplete and suggestions fire on partial input in a search-as-you-type experience, so a RAG pipeline can skip them. Relevance and preparation features, on the other hand, are central to RAG quality.

    Next → 4 / 10

  5. Question 5 of 10

    A RAG filter on a category field returns nothing for 'laptops' although documents contain 'Laptops'. What explains it?

    1. AFilters are skipped unless semantic ranking is on
    2. BFilter text must match verbatim, including case
    3. CFilter text is tokenized like a full-text query
    4. DFilters only work together with vector queries
    Show the answer

    Filters don't go through text analysis. The text in a filter expression is treated as a verbatim, case-sensitive pattern that either matches or fails, so 'laptops' and 'Laptops' are different values.

    Next → 5 / 10

  6. Keep the ones you got wrong

    In the app, every question you miss comes back exactly when you’re about to forget it.

    Get it on Google Play

  7. Question 6 of 10

    In agentic retrieval, which object do you query from code, and what does it unify?

    1. AA search index, unifying remote SharePoint and Bing
    2. BA knowledge source, unifying one or more knowledge bases
    3. CA knowledge base, unifying one or more knowledge sources
    4. DA skillset, unifying the indexers that feed the index
    Show the answer

    Agentic retrieval adds new objects. Knowledge sources each point to searchable content, such as an index or a remote SharePoint site. The knowledge base is the queryable object that unifies them, and your code calls the retrieve action against it.

    Next → 6 / 10

  8. Question 7 of 10

    Which statement about embeddings and cost in a RAG solution is accurate?

    1. AEmbeddings are created only once, when the model is deployed
    2. BVector search needs embeddings at indexing and often at query time
    3. CHybrid search avoids embeddings by using inverted indexes alone
    4. DRetrieved passages reduce input tokens compared with a plain prompt
    Show the answer

    RAG adds work compared with a model-only request. Vector search requires embeddings when content is indexed, and often again for the incoming question, and retrieved passages increase input tokens. Each of these adds cost or latency.

    Next → 7 / 10

  9. Question 8 of 10

    A team is building a new chatbot whose users ask long, conversational, multi-part questions. It wants citations and query details in the response and can accept preview features. Which approach fits?

    1. AFull-text search with fuzzy matching
    2. BClassic RAG with autocomplete enabled
    3. CAgentic retrieval over a knowledge base
    4. DClassic RAG with one hybrid query per turn
    Show the answer

    Classic RAG would also work, which is why this is a real decision. The signals all point one way: a chatbot client, complex conversational questions, citations and query details, and a new implementation. Agentic retrieval plans focused subqueries with conversation history and returns structured responses.

    Next → 8 / 10

  10. Question 9 of 10

    A company allows only generally available features. Its RAG app already has orchestration code, speed matters more than peak relevance, and users phrase things differently from the documents. Which setup fits?

    1. AClassic RAG with a vector-only query
    2. BAgentic retrieval at minimal reasoning effort
    3. CClassic RAG with hybrid queries and semantic ranking
    4. DAgentic retrieval with answer synthesis
    Show the answer

    The GA-only rule and the existing orchestration code point to classic RAG. Within it, hybrid queries combine keyword and vector search for maximum recall, and semantic ranking re-scores results by meaning, which handles the wording mismatch.

    Next → 9 / 10

  11. Question 10 of 10

    HR and finance documents share one Azure AI Search index used for RAG. Finance content must reach only the finance team, and the app is going to production. Which design fits?

    1. AA separate API key for each team on the search service
    2. BA system message telling the model to hide finance data
    3. CCitations on every answer so users can see the sources
    4. DSecurity filters at retrieval time, with Entra ID auth
    Show the answer

    Access must be enforced before content reaches the model. Document-level access control with security filters returns only authorized content, and Microsoft Entra ID is preferred over API keys for production. Retrieved content is untrusted input, so prompts alone don't protect it.

    Next → 10 / 10

  12. You’ve finished this set

    That’s 10 questions on Retrieval and Indexing Methods. In the app the ones you miss come back exactly when you’re about to forget them.

    Back to the course Next lesson: Agent Memory, Tools, and Knowledge Get it on Google Play
Get it on Google Play

The whole course, on your phone

Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.

Get it on Google Play

“This app is the ‘One Lake’ for preparing the certification exams. Really worth it.”

Abdelhak Y. Google Play review