FEFreeExamDumps.in

DP-800 Practice Questions — Page 7

Question 61

Open question ↗

A hybrid search solution must combine product-code precision with semantic troubleshooting help.

Which ranking design should you recommend?

  • A.Combine lexical and vector signals in the ranking process.
  • B.Use semantic similarity only for all searches.
  • C.Remove filters before ranking.
  • D.Sort only by document creation date.

Question 62

Open question ↗

Search quality reviews reveal that older but higher-authority documents should rank above loosely similar recent notes.

Which action should you recommend?

  • A.Add reranking features that account for authority and freshness alongside relevance.
  • B.Ignore metadata once vectors exist.
  • C.Return results in random order.
  • D.Always sort by primary key.

Question 63

Open question ↗

Search results must be filtered by tenant and language before ranking. The solution must prevent documents from other tenants from appearing in the result set.

Which design should you recommend?

  • A.Apply metadata filters before or alongside ranking.
  • B.Rank all documents globally and filter at the client afterward.
  • C.Ignore tenant metadata when embeddings are present.
  • D.Return all matching rows and let the browser hide unauthorized results.

Question 64

Open question ↗

Tenant isolation must be enforced in the search pipeline.

Which action should you recommend?

  • A.Apply tenant filters before returning or reranking candidates.
  • B.Filter only after the user reads the answer.
  • C.Cache unrestricted results for all tenants.
  • D.Disable tenant identifiers in the index.

Question 65

Open question ↗

A RAG assistant answers questions about internal policies stored in Azure SQL Database. If retrieval returns incomplete context, the assistant must avoid confident but unsupported answers.

Which behavior should you recommend?

  • A.Return an answer anyway and rely on the user to detect inaccuracies.
  • B.Instruct the assistant to acknowledge insufficient context and request clarification or additional retrieval.
  • C.Disable citations to shorten responses.
  • D.Always answer with the most common prior response.

Question 66

Open question ↗

A RAG assistant stores evaluation traces to improve future responses.

Which action should you recommend?

  • A.Log retrieval quality, user feedback, and response outcomes for evaluation.
  • B.Delete all traces immediately so nothing can be learned.
  • C.Measure only token count.
  • D.Track no quality signals.

Question 67

Open question ↗

A RAG workflow generates responses from SQL-backed policy documents. Users must be able to trace answers back to retrieved evidence.

Which feature should you recommend?

  • A.Include citations or source references from the retrieved context.
  • B.Hide all supporting evidence to reduce output length.
  • C.Allow unsupported answers when confidence seems high.
  • D.Remove retrieval from the workflow.

Question 68

Open question ↗

A RAG workflow generates responses from SQL-backed policy documents. Users must be able to trace answers back to retrieved evidence.

Which feature should you recommend?

  • A.Include citations or source references from the retrieved context.
  • B.Hide all supporting evidence to reduce output length.
  • C.Allow unsupported answers when confidence seems high.
  • D.Remove retrieval from the workflow.

Question 69

Open question ↗

An enterprise assistant uses row-level security in SQL to protect sensitive records. Retrieved context must never include rows the current user is not allowed to read.

Which requirement is essential?

  • A.Apply row-level access checks before retrieved data is sent to the language model.
  • B.Let the language model infer which rows are sensitive.
  • C.Retrieve all rows and trim them after response generation.
  • D.Cache unrestricted retrieval results for all users.

Question 70

Open question ↗

Feedback shows that some answers are verbose even when only a short policy summary is needed.

Which action should you recommend?

  • A.Tune prompt instructions and output constraints while preserving grounding.
  • B.Disable retrieval to shorten answers.
  • C.Return the full raw document every time.
  • D.Ignore user feedback.