Practice Exams:

Microsoft AI-103 Dumps

EXAM: Developing AI Apps and Agents on Azure
Verified By Experts
Premium Bundle
$49.99

AI-103 Premium ETE File


  • Premium File 187 Questions & Answers
Whats Included:
  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates

Microsoft AI-103 Practice Test 41

Question 400

During a security tabletop exercise, a security engineer must accept a change for rag permission filtering. The issue to be addressed is: A search-backed agent reveals documents outside the user's team. Which concrete verification action demonstrates that the relevant behavior works?

  1. Apply verified per-user filters or secured retrieval identity before synthesis
  2. Retrieve from all teams and rely on model instructions to hide restricted details
  3. Run identical queries as two users with different access grants
  4. Compare labeled query relevance before and after semantic reranking

Explanation

Correct Answer

C:

To verify this change, the large online retailer should run identical queries as two users with different access grants. Operators still need the implementation to apply verified per-user filters or secured retrieval identity before synthesis and the preventive requirement to test deny conditions and enforce ACL trimming at retrieval.

Incorrect Answers

A:

The action to apply verified per-user filters or secured retrieval identity before synthesis can contribute to rag permission filtering as a implementation step; here, a search-backed agent reveals documents outside the user's team. In the large online retailer acceptance review, trying to apply verified per-user filters or secured retrieval identity before synthesis does not test the exact rag permission filtering condition where a search-backed agent reveals documents outside the user's team; the relevant test is to run identical queries as two users with different access grants.

B:

In the large online retailer acceptance review, trying to retrieve from all teams and rely on model instructions to hide restricted details does not test the exact rag permission filtering condition where a search-backed agent reveals documents outside the user's team; the relevant test is to run identical queries as two users with different access grants. Acceptance should instead demonstrate that the service can run identical queries as two users with different access grants, using the described failure as the test criterion.

D:

In the large online retailer acceptance review, trying to compare labeled query relevance before and after semantic reranking does not test the exact rag permission filtering condition where a search-backed agent reveals documents outside the user's team; the relevant test is to run identical queries as two users with different access grants. Without the actual change to run identical queries as two users with different access grants, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

Question 401

At a financial services division, the agent designer is preparing a design meeting. The concrete requirement is: A model answers from low-relevance chunks despite no suitable document match. Which design action BEST addresses the stated need?

  1. Define retrieval confidence and no-answer behavior using evaluated thresholds
  2. Ask questions absent from the corpus and confirm grounded uncertainty
  3. Update prompt and tool schemas together and validate supported versions
  4. Treat every vector neighbor as relevant just because one was returned

Explanation

Correct Answer

A:

The proposed implementation is to define retrieval confidence and no-answer behavior using evaluated thresholds; this matches the operational requirement that a model answers from low-relevance chunks despite no suitable document match. The team can ask questions absent from the corpus and confirm grounded uncertainty. The team should also evaluate false-answer risk with known unanswerable questions.

Incorrect Answers

B:

The action to ask questions absent from the corpus and confirm grounded uncertainty can contribute to search score thresholds as a targeted acceptance check step; here, a model answers from low-relevance chunks despite no suitable document match. Choosing to ask questions absent from the corpus and confirm grounded uncertainty cannot directly replace the specified action to define retrieval confidence and no-answer behavior using evaluated thresholds when a model answers from low-relevance chunks despite no suitable document match.

C:

Choosing to update prompt and tool schemas together and validate supported versions would direct effort toward conversation tool schema migration, whose characteristic issue is that an application tool was renamed and the agent still emits the removed argument. The relevant corrective or preventive action here remains to define retrieval confidence and no-answer behavior using evaluated thresholds, with its own validation criteria.

D:

Choosing to treat every vector neighbor as relevant just because one was returned cannot directly replace the specified action to define retrieval confidence and no-answer behavior using evaluated thresholds when a model answers from low-relevance chunks despite no suitable document match. The actual implementation must define retrieval confidence and no-answer behavior using evaluated thresholds, because the stated service condition is that a model answers from low-relevance chunks despite no suitable document match.

Question 402

After a staged service rollout at a logistics operator, the team observes this failure: A model answers from low-relevance chunks despite no suitable document match. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Treat every vector neighbor as relevant just because one was returned
  2. Truncate the end of the document and assume no key requirements are lost
  3. Define retrieval confidence and no-answer behavior using evaluated thresholds
  4. Ask questions absent from the corpus and confirm grounded uncertainty

Explanation

Correct Answer

A:

This design mistake is to treat every vector neighbor as relevant just because one was returned; it conflicts with the actual search score thresholds behavior needed when a model answers from low-relevance chunks despite no suitable document match. The logistics operator should instead define retrieval confidence and no-answer behavior using evaluated thresholds, followed by a check to ask questions absent from the corpus and confirm grounded uncertainty.

Incorrect Answers

B:

The option to truncate the end of the document and assume no key requirements are lost would normally serve a different long document summarization requirement, where a lengthy policy manual exceeds the model context window. The alternative may be useful in its own context, but it does not directly supply treat every vector neighbor as relevant just because one was returned for the stated decision.

C:

The alternative to define retrieval confidence and no-answer behavior using evaluated thresholds is a implementation step rather than the faulty assumption being diagnosed. This logistics operator failure-analysis alternative of define retrieval confidence and no-answer behavior using evaluated thresholds does not diagnose the particular search score thresholds defect where a model answers from low-relevance chunks despite no suitable document match; the keyed mistake is treat every vector neighbor as relevant just because one was returned.

D:

The alternative to ask questions absent from the corpus and confirm grounded uncertainty is a targeted acceptance check step rather than the faulty assumption being diagnosed. This logistics operator failure-analysis alternative of ask questions absent from the corpus and confirm grounded uncertainty does not diagnose the particular search score thresholds defect where a model answers from low-relevance chunks despite no suitable document match; the keyed mistake is treat every vector neighbor as relevant just because one was returned.

Question 403

A solution architect at a energy cooperative is conducting a release acceptance review concerning search score thresholds. The observed problem is: A model answers from low-relevance chunks despite no suitable document match. Which TWO actions together represent the needed implementation AND a durable preventive safeguard? (Select TWO answers.)

  1. Map generated claims to specific retrieved evidence spans and reject unsupported claims
  2. Evaluate false-answer risk with known unanswerable questions
  3. Ask questions absent from the corpus and confirm grounded uncertainty
  4. Define retrieval confidence and no-answer behavior using evaluated thresholds
  5. Treat every vector neighbor as relevant just because one was returned

Explanation

Correct Answers

B:

The sustained prevention measure is to evaluate false-answer risk with known unanswerable questions. The implementation can still use the separate action to define retrieval confidence and no-answer behavior using evaluated thresholds, but a guardrail constrains future changes. The energy cooperative can measure whether the rule works by arranging to ask questions absent from the corpus and confirm grounded uncertainty.

D:

The proposed implementation is to define retrieval confidence and no-answer behavior using evaluated thresholds; this matches the operational requirement that a model answers from low-relevance chunks despite no suitable document match. The team can ask questions absent from the corpus and confirm grounded uncertainty. The team should also evaluate false-answer risk with known unanswerable questions.

Incorrect Answers

A:

The action to map generated claims to specific retrieved evidence spans and reject unsupported claims addresses citation alignment rather than the decisive search score thresholds condition. The energy cooperative should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to define retrieval confidence and no-answer behavior using evaluated thresholds.

C:

The action to ask questions absent from the corpus and confirm grounded uncertainty can contribute to search score thresholds as a targeted acceptance check step; here, a model answers from low-relevance chunks despite no suitable document match. During energy cooperative remediation, the proposed step to ask questions absent from the corpus and confirm grounded uncertainty is not the direct restoration action for search score thresholds when a model answers from low-relevance chunks despite no suitable document match; the keyed repair is evaluate false-answer risk with known unanswerable questions.

E:

During energy cooperative remediation, the proposed step to treat every vector neighbor as relevant just because one was returned is not the direct restoration action for search score thresholds when a model answers from low-relevance chunks despite no suitable document match; the keyed repair is evaluate false-answer risk with known unanswerable questions. After remediation the team must ask questions absent from the corpus and confirm grounded uncertainty; the action in this option does not itself establish that target recovery outcome.

Question 404

A machine-learning developer at a industrial maintenance team wants the following issue to be prevented in later deployments: A model answers from low-relevance chunks despite no suitable document match. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Evaluate false-answer risk with known unanswerable questions
  2. Require policy citations and reject unsupported high-impact claims
  3. Treat every vector neighbor as relevant just because one was returned
  4. Apply calibrated relevance gates and prompt abstention when evidence is weak

Explanation

Correct Answer

A:

Standardizing the boundary to evaluate false-answer risk with known unanswerable questions is preventive because the same search score thresholds failure could otherwise return: a model answers from low-relevance chunks despite no suitable document match. For this industrial maintenance team, the service should also define retrieval confidence and no-answer behavior using evaluated thresholds, and tests should ask questions absent from the corpus and confirm grounded uncertainty so that the ongoing requirement is measurable.

Incorrect Answers

B:

The action to require policy citations and reject unsupported high-impact claims addresses retrieval-augmented policy answers rather than the decisive search score thresholds condition. The industrial maintenance team should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to evaluate false-answer risk with known unanswerable questions.

C:

For prevention at the industrial maintenance team, the measure to treat every vector neighbor as relevant just because one was returned is weaker or aimed at another issue than the search score thresholds guardrail to evaluate false-answer risk with known unanswerable questions needed when a model answers from low-relevance chunks despite no suitable document match. The lasting guardrail for this specific threat is to evaluate false-answer risk with known unanswerable questions, not simply to complete an adjacent service task.

D:

The action to apply calibrated relevance gates and prompt abstention when evidence is weak can contribute to search score thresholds as a corrective remediation step; here, a model answers from low-relevance chunks despite no suitable document match. For prevention at the industrial maintenance team, the measure to apply calibrated relevance gates and prompt abstention when evidence is weak is weaker or aimed at another issue than the search score thresholds guardrail to evaluate false-answer risk with known unanswerable questions needed when a model answers from low-relevance chunks despite no suitable document match.

Question 405

During a staged service rollout, a AI application owner must accept a change for search score thresholds. The issue to be addressed is: A model answers from low-relevance chunks despite no suitable document match. Which concrete verification action demonstrates that the relevant behavior works?

  1. Ask questions absent from the corpus and confirm grounded uncertainty
  2. Treat every vector neighbor as relevant just because one was returned
  3. Run supported and unsupported test questions and score factual support
  4. Define retrieval confidence and no-answer behavior using evaluated thresholds

Explanation

Correct Answer

A:

The targeted acceptance test is to ask questions absent from the corpus and confirm grounded uncertainty. The design should first define retrieval confidence and no-answer behavior using evaluated thresholds. A robust research university deployment also needs to evaluate false-answer risk with known unanswerable questions; passing this check provides relevant evidence but does not certify every possible input or security risk.

Incorrect Answers

B:

In the research university acceptance review, trying to treat every vector neighbor as relevant just because one was returned does not test the exact search score thresholds condition where a model answers from low-relevance chunks despite no suitable document match; the relevant test is to ask questions absent from the corpus and confirm grounded uncertainty. Acceptance should instead demonstrate that the service can ask questions absent from the corpus and confirm grounded uncertainty, using the described failure as the test criterion.

C:

For model evaluation for fabrication, it can be sensible to run supported and unsupported test questions and score factual support if a new retrieval prompt sounds fluent but invents technical product restrictions. Without the actual change to ask questions absent from the corpus and confirm grounded uncertainty, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

D:

The action to define retrieval confidence and no-answer behavior using evaluated thresholds can contribute to search score thresholds as a implementation step; here, a model answers from low-relevance chunks despite no suitable document match. In the research university acceptance review, trying to define retrieval confidence and no-answer behavior using evaluated thresholds does not test the exact search score thresholds condition where a model answers from low-relevance chunks despite no suitable document match; the relevant test is to ask questions absent from the corpus and confirm grounded uncertainty.

Question 406

At a large online retailer, the quality assurance lead is preparing a data protection inspection. The concrete requirement is: A sales report assistant confuses figures across table headers. Which design action BEST addresses the stated need?

  1. Evaluate factual groundedness using source-supported queries and adversarial missing-fact cases
  2. Use layout-aware parsing and explicit column-row associations for table extraction
  3. Flatten rows into unrelated word sequences and ignore header positions
  4. Ask for a known row-column intersection and compare the returned value

Explanation

Correct Answer

B:

Choosing to use layout-aware parsing and explicit column-row associations for table extraction directly targets the described grounded table question answering gap: a sales report assistant confuses figures across table headers. As a lasting boundary, they also need to test numeric joins and citations to the correct table cell, because an implementation by itself does not prove safe operation.

Incorrect Answers

A:

Choosing to evaluate factual groundedness using source-supported queries and adversarial missing-fact cases would direct effort toward model evaluation for fabrication, whose characteristic issue is that a new retrieval prompt sounds fluent but invents technical product restrictions. The relevant corrective or preventive action here remains to use layout-aware parsing and explicit column-row associations for table extraction, with its own validation criteria.

C:

Choosing to flatten rows into unrelated word sequences and ignore header positions cannot directly replace the specified action to use layout-aware parsing and explicit column-row associations for table extraction when a sales report assistant confuses figures across table headers. The actual implementation must use layout-aware parsing and explicit column-row associations for table extraction, because the stated service condition is that a sales report assistant confuses figures across table headers.

D:

The action to ask for a known row-column intersection and compare the returned value can contribute to grounded table question answering as a targeted acceptance check step; here, a sales report assistant confuses figures across table headers. Choosing to ask for a known row-column intersection and compare the returned value cannot directly replace the specified action to use layout-aware parsing and explicit column-row associations for table extraction when a sales report assistant confuses figures across table headers.

Question 407

After a quality regression test at a public library network, the team observes this failure: A sales report assistant confuses figures across table headers. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Use layout-aware parsing and explicit column-row associations for table extraction
  2. Summarize whichever document appears first without checking applicability
  3. Test numeric joins and citations to the correct table cell
  4. Flatten rows into unrelated word sequences and ignore header positions
  5. Ask for a known row-column intersection and compare the returned value

Explanation

Correct Answer

D:

The faulty assumption is flatten rows into unrelated word sequences and ignore header positions. A functioning alternative would be to use layout-aware parsing and explicit column-row associations for table extraction, and the relevant validation would be to ask for a known row-column intersection and compare the returned value.

Incorrect Answers

A:

The alternative to use layout-aware parsing and explicit column-row associations for table extraction is a implementation step rather than the faulty assumption being diagnosed. This public library network failure-analysis alternative of use layout-aware parsing and explicit column-row associations for table extraction does not diagnose the particular grounded table question answering defect where a sales report assistant confuses figures across table headers; the keyed mistake is flatten rows into unrelated word sequences and ignore header positions.

B:

This approach relates to cross-document synthesis: a compliance answer must reconcile two approved documents with different effective dates. If the team chose it instead, the targeted check to ask for a known row-column intersection and compare the returned value could still fail because the underlying required action is to flatten rows into unrelated word sequences and ignore header positions.

C:

The alternative to test numeric joins and citations to the correct table cell is a ongoing preventive rule step rather than the faulty assumption being diagnosed. This public library network failure-analysis alternative of test numeric joins and citations to the correct table cell does not diagnose the particular grounded table question answering defect where a sales report assistant confuses figures across table headers; the keyed mistake is flatten rows into unrelated word sequences and ignore header positions.

E:

The alternative to ask for a known row-column intersection and compare the returned value is a targeted acceptance check step rather than the faulty assumption being diagnosed. This public library network failure-analysis alternative of ask for a known row-column intersection and compare the returned value does not diagnose the particular grounded table question answering defect where a sales report assistant confuses figures across table headers; the keyed mistake is flatten rows into unrelated word sequences and ignore header positions.

Question 408

The software consultancy is responding to a production incident. A compliance specialist discovers this issue: A sales report assistant confuses figures across table headers. What corrective step should the team take to restore the intended functionality most directly?

  1. Flatten rows into unrelated word sequences and ignore header positions
  2. Reingest tables preserving structure and regenerate the disputed answer
  3. Restore compatible tool bindings and replay the affected function call
  4. Ask for a known row-column intersection and compare the returned value
  5. Test numeric joins and citations to the correct table cell
  6. Use layout-aware parsing and explicit column-row associations for table extraction

Explanation

Correct Answer

B:

The appropriate remediation is to reingest tables preserving structure and regenerate the disputed answer, because the failure concerns grounded table question answering: a sales report assistant confuses figures across table headers. A meaningful verification should ask for a known row-column intersection and compare the returned value, and the durable operating rule should test numeric joins and citations to the correct table cell after the immediate fix.

Incorrect Answers

A:

During software consultancy remediation, the proposed step to flatten rows into unrelated word sequences and ignore header positions is not the direct restoration action for grounded table question answering when a sales report assistant confuses figures across table headers; the keyed repair is reingest tables preserving structure and regenerate the disputed answer. After remediation the team must ask for a known row-column intersection and compare the returned value; the action in this option does not itself establish that target recovery outcome.

C:

For conversation tool schema migration, it can be sensible to restore compatible tool bindings and replay the affected function call if an application tool was renamed and the agent still emits the removed argument. Without the actual change to reingest tables preserving structure and regenerate the disputed answer, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

D:

The action to ask for a known row-column intersection and compare the returned value can contribute to grounded table question answering as a targeted acceptance check step; here, a sales report assistant confuses figures across table headers. During software consultancy remediation, the proposed step to ask for a known row-column intersection and compare the returned value is not the direct restoration action for grounded table question answering when a sales report assistant confuses figures across table headers; the keyed repair is reingest tables preserving structure and regenerate the disputed answer.

E:

The action to test numeric joins and citations to the correct table cell can contribute to grounded table question answering as a ongoing preventive rule step; here, a sales report assistant confuses figures across table headers. During software consultancy remediation, the proposed step to test numeric joins and citations to the correct table cell is not the direct restoration action for grounded table question answering when a sales report assistant confuses figures across table headers; the keyed repair is reingest tables preserving structure and regenerate the disputed answer.

F:

The action to use layout-aware parsing and explicit column-row associations for table extraction can contribute to grounded table question answering as a implementation step; here, a sales report assistant confuses figures across table headers. During software consultancy remediation, the proposed step to use layout-aware parsing and explicit column-row associations for table extraction is not the direct restoration action for grounded table question answering when a sales report assistant confuses figures across table headers; the keyed repair is reingest tables preserving structure and regenerate the disputed answer.

Question 409

A product support lead at a regional hospital wants the following issue to be prevented in later deployments: A sales report assistant confuses figures across table headers. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Test numeric joins and citations to the correct table cell
  2. Reingest tables preserving structure and regenerate the disputed answer
  3. Flatten rows into unrelated word sequences and ignore header positions
  4. Preserve tenant session boundaries and limit unnecessary history

Explanation

Correct Answer

A:

The sustained prevention measure is to test numeric joins and citations to the correct table cell. The implementation can still use the separate action to use layout-aware parsing and explicit column-row associations for table extraction, but a guardrail constrains future changes. The regional hospital can measure whether the rule works by arranging to ask for a known row-column intersection and compare the returned value.

Incorrect Answers

B:

The action to reingest tables preserving structure and regenerate the disputed answer can contribute to grounded table question answering as a corrective remediation step; here, a sales report assistant confuses figures across table headers. For prevention at the regional hospital, the measure to reingest tables preserving structure and regenerate the disputed answer is weaker or aimed at another issue than the grounded table question answering guardrail to test numeric joins and citations to the correct table cell needed when a sales report assistant confuses figures across table headers.

C:

For prevention at the regional hospital, the measure to flatten rows into unrelated word sequences and ignore header positions is weaker or aimed at another issue than the grounded table question answering guardrail to test numeric joins and citations to the correct table cell needed when a sales report assistant confuses figures across table headers. The lasting guardrail for this specific threat is to test numeric joins and citations to the correct table cell, not simply to complete an adjacent service task.

D:

The action to preserve tenant session boundaries and limit unnecessary history addresses multi-turn rag grounding rather than the decisive grounded table question answering condition. The regional hospital should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to test numeric joins and citations to the correct table cell.

Question 410

During a quality regression test, a data engineer must accept a change for grounded table question answering. The issue to be addressed is: A sales report assistant confuses figures across table headers. Which concrete verification action demonstrates that the relevant behavior works?

  1. Ask a follow-up about a named earlier policy and inspect retrieved passages
  2. Use layout-aware parsing and explicit column-row associations for table extraction
  3. Flatten rows into unrelated word sequences and ignore header positions
  4. Ask for a known row-column intersection and compare the returned value

Explanation

Correct Answer

D:

To verify this change, the insurance platform should ask for a known row-column intersection and compare the returned value. Operators still need the implementation to use layout-aware parsing and explicit column-row associations for table extraction and the preventive requirement to test numeric joins and citations to the correct table cell.

Incorrect Answers

A:

For multi-turn rag grounding, it can be sensible to ask a follow-up about a named earlier policy and inspect retrieved passages if a user asks a follow-up whose meaning depends on prior context. Without the actual change to ask for a known row-column intersection and compare the returned value, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

B:

The action to use layout-aware parsing and explicit column-row associations for table extraction can contribute to grounded table question answering as a implementation step; here, a sales report assistant confuses figures across table headers. In the insurance platform acceptance review, trying to use layout-aware parsing and explicit column-row associations for table extraction does not test the exact grounded table question answering condition where a sales report assistant confuses figures across table headers; the relevant test is to ask for a known row-column intersection and compare the returned value.

C:

In the insurance platform acceptance review, trying to flatten rows into unrelated word sequences and ignore header positions does not test the exact grounded table question answering condition where a sales report assistant confuses figures across table headers; the relevant test is to ask for a known row-column intersection and compare the returned value. Acceptance should instead demonstrate that the service can ask for a known row-column intersection and compare the returned value, using the described failure as the test criterion.

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