McAfee-Secured Website

Exam Code: AI-103

Exam Name: Developing AI Apps and Agents on Azure

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Apps and Agents Developer Associate

100% Microsoft AI-103 Guaranteed Success With Testking Training Materials

Get Real AI-103 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

187 Questions & Answers with Testing Engine

"Developing AI Apps and Agents on Azure Exam", also known as AI-103 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-103 Exam Engine. Your AI-103 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-103 Practice Test 19

Question 180

An inference endpoint rejects a harmful user request with a content-filter outcome. A developer plans to return the rejected raw prompt in a detailed error page. What should the app do?

  1. Retry the same blocked request indefinitely until one attempt gets through.
  2. Handle content filter responses without returning unsanitized raw text.
  3. Disable moderation for users whose initial request was rejected.
  4. Treat the filter error as a successful generated answer and display its entire payload.

Explanation

Correct Answer

B:

Safety interventions should be surfaced without reintroducing prohibited material through the UI.

Incorrect Answers

A:

Unlimited retries undermine policy enforcement and waste resources.

C:

A rejection is not authorization to remove safety controls.

D:

This can bypass the intended safety treatment of the rejected content.

Question 181

A deployed model intermittently returns 503. An alternate validated deployment exists. What resilience behavior is appropriate?

  1. Return a fake completed transaction when inference is unavailable.
  2. Repeat requests in an unbounded tight loop on the same failing deployment.
  3. Apply bounded retries/backoff and a policy-approved fallback with clear failure signaling.
  4. Increase the maximum completion-token setting while leaving transient capacity errors and retry behavior untouched.

Explanation

Correct Answer

C:

Transient capacity failures can be retried or routed to a compatible backup without fabricating results.

Incorrect Answers

A:

Inventing success conceals a real failure from downstream callers.

B:

Aggressive retry storms can worsen temporary service disruption.

D:

Token-output limits control answer length, not temporary service availability or permitted fallback handling.

Question 182

An agent's payment submission times out after the gateway may already have accepted it. What retry rule avoids duplicate charges?

  1. Use a healthy API gateway status as proof a specific financial transaction has been committed.
  2. Change the model's completion temperature and resend the write.
  3. Immediately repeat the payment request without checking the previous outcome.
  4. Avoid automatic retries on non-idempotent downstream operations.

Explanation

Correct Answer

D:

Payment creation may have side effects even if the original response was lost.

Incorrect Answers

A:

Gateway uptime is unrelated to the state of an individual non-idempotent transaction.

B:

Sampling settings do not guarantee idempotent business transactions.

C:

Blind retries can charge the customer twice.

Question 183

A research archive needs semantic retrieval for paragraphs describing the same experiment with different vocabulary. What must be indexed?

  1. Embed internal paragraphs for semantic document retrieval.
  2. Use the chat-completion deployment's display name as an embedding array.
  3. Store document page counts as the only vector values.
  4. Convert all paragraphs into timestamps before indexing.

Explanation

Correct Answer

A:

A shared embedding representation supports meaning-based nearest-neighbor retrieval.

Incorrect Answers

B:

An endpoint identifier is not a numerical document representation.

C:

Page counts do not encode the scientific meaning of the passage.

D:

Timestamps are metadata and do not capture semantic similarity.

Question 184

A compliance answer cites a contract requirement that has a nearby exception. Naive prompt assembly separates the exception from its clause. What should change?

  1. Split each contract at an arbitrary character count and discard overlaps.
  2. Keep the contract clause and its exception in the same cited context block.
  3. Sort all contract sentences alphabetically before prompt assembly.
  4. Use only the contract heading in the response prompt.

Explanation

Correct Answer

B:

Preserving local context prevents the model from applying a requirement without its qualifying condition.

Incorrect Answers

A:

Mechanical slicing can separate conditions needed for correct interpretation.

C:

Reordering destroys legal relationships and chronology.

D:

A heading lacks substantive evidence for the exception.

Question 185

An indexed handbook stores useful semantic vectors, but the chatbot cannot link cited passages back to source pages. What should indexing preserve?

  1. Issue fresh display citation labels for every query without preserving their backing document IDs.
  2. Store document ID and chunk offsets alongside embeddings.
  3. Record only the embedding model temperature for each passage.
  4. Store vectors without any record of their original documents.

Explanation

Correct Answer

B:

Source metadata allows retrieved fragments to be traced and cited independently of model generation.

Incorrect Answers

A:

A user-friendly label must resolve to stable real source metadata for auditability.

C:

Generation sampling does not identify a chunk's document or location.

D:

Unattributed vectors cannot produce reliable source references.

Question 186

An assistant claims 'according to policy page 19' but the retrieved search result points to a different document. How should citations be produced?

  1. Return search relevance scores as clickable document paths.
  2. Retrieve citations from indexed source metadata in generated answers.
  3. Use the user query as the citation URL.
  4. Let the model synthesize plausible page labels without consulting retrieved metadata.

Explanation

Correct Answer

B:

Deterministic source references avoid inventing page numbers in generated prose.

Incorrect Answers

A:

Similarity scores are not document identifiers or URLs.

C:

A question is not a link to supporting policy evidence.

D:

Fluent generated locators can misrepresent the source of a claim.

Question 187

A RAG chatbot serves HR policy documents. A user may see company-wide guidance but not another employee's salary letter. What must retrieval enforce?

  1. Search globally and instruct the model to conceal restricted content afterward.
  2. Expose complete search results only to the browser and ask users to ignore irrelevant private letters.
  3. Enforce document ACL filters before retrieved passages reach the model.
  4. Use similarity-based passage ranking as a proxy for user authorization without document ACL filters.

Explanation

Correct Answer

C:

Permission-aware retrieval prevents unauthorized evidence from being surfaced by later generation.

Incorrect Answers

A:

Sensitive text may already be read and logged by the model before redaction.

B:

UI masking is not an adequate document authorization boundary.

D:

Search relevance does not imply entitlement to an employee’s confidential documents.

Question 188

A contracts search must match exact article code SEC-18 and also retrieve sentences paraphrasing 'renewal failure.' What ranking setup fits?

  1. Use solely vector similarity and ignore the literal article code.
  2. Use solely full-text token matching for all possible paraphrases.
  3. Avoid retrieval and ask the generative model to recall proprietary contract codes.
  4. Use hybrid search for a question with exact legal clause codes and synonyms.

Explanation

Correct Answer

D:

The lexical branch preserves exact identifiers while the vector branch finds semantic alternatives.

Incorrect Answers

A:

A nearest semantic result might reference a different article code.

B:

Lexical matching can miss relevant text phrased with different terminology.

C:

Private clause facts should come from the indexed source, not model memory.

Question 189

Hybrid search returns the right policy somewhere in the first 50 results but the top few are often weak. Which optional stage can improve the ordering?

  1. Increase output tokens because a longer answer automatically reranks evidence.
  2. Prioritize document age or size as the only ranking signal rather than semantic relevance to the current query.
  3. Semantically rerank the hybrid search candidate set.
  4. Delete the initial retrieval step and ask the ranker to search the entire corpus from nothing.

Explanation

Correct Answer

C:

Reranking refines candidate relevance after initial lexical/vector recall.

Incorrect Answers

A:

Generation length does not reorder search results.

B:

Recency and size may be useful metadata, but they do not replace task-relevant semantic ranking.

D:

A reranker operates on retrieved candidates and is not a complete indexing substitute.

Question 190

A RAG model receives an empty `context` block for a newly published FAQ, despite the upstream source containing the article. Where should diagnosis begin?

  1. Remove all source logging because no context was found.
  2. Increase model temperature and ask it to reconstruct the missing FAQ.
  3. Replace the user question with a generic instruction to be helpful.
  4. Diagnose an empty context from a stale search index.

Explanation

Correct Answer

D:

Missing indexed content or overrestrictive filters can explain an empty retrieval context.

Incorrect Answers

A:

Suppressing diagnostics makes freshness and filter failures harder to investigate.

B:

Sampling creativity cannot access the absent authoritative source.

C:

A vaguer question will not prove the new article is indexed.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests