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 103

Question 1020

During a cost review, a machine-learning developer must accept a change for search cache staleness. The issue to be addressed is: A user sees outdated answers even though the index contains new records. Which concrete verification action demonstrates that the relevant behavior works?

  1. Assume a successful indexer refresh invalidates every external app cache
  2. Query a known API name and a paraphrased failure in the same test suite
  3. Evict stale cache records and verify retrieval from the new index state
  4. Change a known document and check downstream cached retrieval results
  5. Invalidate application retrieval caches keyed to obsolete document versions
  6. Attach source revision and expiry to cached answers

Explanation

Correct Answer

D:

To verify this change, the customer support center should change a known document and check downstream cached retrieval results. Operators still need the implementation to invalidate application retrieval caches keyed to obsolete document versions and the preventive requirement to attach source revision and expiry to cached answers.

Incorrect Answers

A:

In the customer support center acceptance review, trying to assume a successful indexer refresh invalidates every external app cache does not test the exact search cache staleness condition where a user sees outdated answers even though the index contains new records; the relevant test is to change a known document and check downstream cached retrieval results. Acceptance should instead demonstrate that the service can change a known document and check downstream cached retrieval results, using the described failure as the test criterion.

B:

The option to query a known API name and a paraphrased failure in the same test suite would normally serve a different hybrid search fusion requirement, where users need both exact API identifiers and conceptual troubleshooting guidance. The alternative may be useful in its own context, but it does not directly supply change a known document and check downstream cached retrieval results for the stated decision.

C:

The action to evict stale cache records and verify retrieval from the new index state can contribute to search cache staleness as a corrective remediation step; here, a user sees outdated answers even though the index contains new records. In the customer support center acceptance review, trying to evict stale cache records and verify retrieval from the new index state does not test the exact search cache staleness condition where a user sees outdated answers even though the index contains new records; the relevant test is to change a known document and check downstream cached retrieval results.

E:

The action to invalidate application retrieval caches keyed to obsolete document versions can contribute to search cache staleness as a implementation step; here, a user sees outdated answers even though the index contains new records. In the customer support center acceptance review, trying to invalidate application retrieval caches keyed to obsolete document versions does not test the exact search cache staleness condition where a user sees outdated answers even though the index contains new records; the relevant test is to change a known document and check downstream cached retrieval results.

F:

The action to attach source revision and expiry to cached answers can contribute to search cache staleness as a ongoing preventive rule step; here, a user sees outdated answers even though the index contains new records. In the customer support center acceptance review, trying to attach source revision and expiry to cached answers does not test the exact search cache staleness condition where a user sees outdated answers even though the index contains new records; the relevant test is to change a known document and check downstream cached retrieval results.

Question 1021

At a pharmaceutical research group, the AI application owner is preparing a support escalation. The concrete requirement is: A handbook contains screenshots whose meaning is absent from extracted text. Which design action BEST addresses the stated need?

  1. Ask about a labeled screenshot component and check image-derived evidence
  2. Return stable document and chunk identifiers with retrieval and generation logs
  3. Discard all images and claim diagrams are fully captured by text-only indexing
  4. Use supported visual extraction or image verbalization enrichment before indexing

Explanation

Correct Answer

D:

The proposed implementation is to use supported visual extraction or image verbalization enrichment before indexing; this matches the operational requirement that a handbook contains screenshots whose meaning is absent from extracted text. The team can ask about a labeled screenshot component and check image-derived evidence. The team should also retain visual source coordinates and evaluate critical diagram recall.

Incorrect Answers

A:

The action to ask about a labeled screenshot component and check image-derived evidence can contribute to multimodal indexing as a targeted acceptance check step; here, a handbook contains screenshots whose meaning is absent from extracted text. Choosing to ask about a labeled screenshot component and check image-derived evidence cannot directly replace the specified action to use supported visual extraction or image verbalization enrichment before indexing when a handbook contains screenshots whose meaning is absent from extracted text.

B:

Choosing to return stable document and chunk identifiers with retrieval and generation logs would direct effort toward grounding metadata trace, whose characteristic issue is that an agent response has a citation URL but no document version or chunk ID. The relevant corrective or preventive action here remains to use supported visual extraction or image verbalization enrichment before indexing, with its own validation criteria.

C:

Choosing to discard all images and claim diagrams are fully captured by text-only indexing cannot directly replace the specified action to use supported visual extraction or image verbalization enrichment before indexing when a handbook contains screenshots whose meaning is absent from extracted text. The actual implementation must use supported visual extraction or image verbalization enrichment before indexing, because the stated service condition is that a handbook contains screenshots whose meaning is absent from extracted text.

Question 1022

After a service audit at a travel support provider, the team observes this failure: A handbook contains screenshots whose meaning is absent from extracted text. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Ask about a labeled screenshot component and check image-derived evidence
  2. Discard all images and claim diagrams are fully captured by text-only indexing
  3. Use supported visual extraction or image verbalization enrichment before indexing
  4. Expect a field that was not configured as filterable to support arbitrary filters

Explanation

Correct Answer

B:

This design mistake is to discard all images and claim diagrams are fully captured by text-only indexing; it conflicts with the actual multimodal indexing behavior needed when a handbook contains screenshots whose meaning is absent from extracted text. The travel support provider should instead use supported visual extraction or image verbalization enrichment before indexing, followed by a check to ask about a labeled screenshot component and check image-derived evidence.

Incorrect Answers

A:

The alternative to ask about a labeled screenshot component and check image-derived evidence is a targeted acceptance check step rather than the faulty assumption being diagnosed. This travel support provider failure-analysis alternative of ask about a labeled screenshot component and check image-derived evidence does not diagnose the particular multimodal indexing defect where a handbook contains screenshots whose meaning is absent from extracted text; the keyed mistake is discard all images and claim diagrams are fully captured by text-only indexing.

C:

The alternative to use supported visual extraction or image verbalization enrichment before indexing is a implementation step rather than the faulty assumption being diagnosed. This travel support provider failure-analysis alternative of use supported visual extraction or image verbalization enrichment before indexing does not diagnose the particular multimodal indexing defect where a handbook contains screenshots whose meaning is absent from extracted text; the keyed mistake is discard all images and claim diagrams are fully captured by text-only indexing.

D:

This travel support provider failure-analysis alternative of expect a field that was not configured as filterable to support arbitrary filters does not diagnose the particular multimodal indexing defect where a handbook contains screenshots whose meaning is absent from extracted text; the keyed mistake is discard all images and claim diagrams are fully captured by text-only indexing. The alternative may be useful in its own context, but it does not directly supply discard all images and claim diagrams are fully captured by text-only indexing for the stated decision.

Question 1023

A AI platform engineer at a scientific publisher is conducting a security tabletop exercise concerning multimodal indexing. The observed problem is: A handbook contains screenshots whose meaning is absent from extracted text. Which TWO actions together represent the needed implementation AND a durable preventive safeguard? (Select TWO answers.)

  1. Use supported visual extraction or image verbalization enrichment before indexing
  2. Ask about a labeled screenshot component and check image-derived evidence
  3. Retain visual source coordinates and evaluate critical diagram recall
  4. Discard all images and claim diagrams are fully captured by text-only indexing
  5. Run keyword and vector searches in a coordinated hybrid query

Explanation

Correct Answers

A:

The proposed implementation is to use supported visual extraction or image verbalization enrichment before indexing; this matches the operational requirement that a handbook contains screenshots whose meaning is absent from extracted text. The team can ask about a labeled screenshot component and check image-derived evidence. The team should also retain visual source coordinates and evaluate critical diagram recall.

C:

The sustained prevention measure is to retain visual source coordinates and evaluate critical diagram recall. The implementation can still use the separate action to use supported visual extraction or image verbalization enrichment before indexing, but a guardrail constrains future changes. The scientific publisher can measure whether the rule works by arranging to ask about a labeled screenshot component and check image-derived evidence.

Incorrect Answers

B:

The action to ask about a labeled screenshot component and check image-derived evidence can contribute to multimodal indexing as a targeted acceptance check step; here, a handbook contains screenshots whose meaning is absent from extracted text. During scientific publisher remediation, the proposed step to ask about a labeled screenshot component and check image-derived evidence is not the direct restoration action for multimodal indexing when a handbook contains screenshots whose meaning is absent from extracted text; the keyed repair is use supported visual extraction or image verbalization enrichment before indexing.

D:

During scientific publisher remediation, the proposed step to discard all images and claim diagrams are fully captured by text-only indexing is not the direct restoration action for multimodal indexing when a handbook contains screenshots whose meaning is absent from extracted text; the keyed repair is use supported visual extraction or image verbalization enrichment before indexing. After remediation the team must ask about a labeled screenshot component and check image-derived evidence; the action in this option does not itself establish that target recovery outcome.

E:

The action to run keyword and vector searches in a coordinated hybrid query addresses hybrid search fusion rather than the decisive multimodal indexing condition. The scientific publisher should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to use supported visual extraction or image verbalization enrichment before indexing.

Question 1024

A compliance specialist at a nonprofit organization wants the following issue to be prevented in later deployments: A handbook contains screenshots whose meaning is absent from extracted text. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Reprocess diagrams into grounded searchable descriptions
  2. Discard all images and claim diagrams are fully captured by text-only indexing
  3. Maintain evaluation for both exact and semantic intents
  4. Retain visual source coordinates and evaluate critical diagram recall

Explanation

Correct Answer

D:

Standardizing the boundary to retain visual source coordinates and evaluate critical diagram recall is preventive because the same multimodal indexing failure could otherwise return: a handbook contains screenshots whose meaning is absent from extracted text. For this nonprofit organization, the service should also use supported visual extraction or image verbalization enrichment before indexing, and tests should ask about a labeled screenshot component and check image-derived evidence so that the ongoing requirement is measurable.

Incorrect Answers

A:

The action to reprocess diagrams into grounded searchable descriptions can contribute to multimodal indexing as a corrective remediation step; here, a handbook contains screenshots whose meaning is absent from extracted text. For prevention at the nonprofit organization, the measure to reprocess diagrams into grounded searchable descriptions is weaker or aimed at another issue than the multimodal indexing guardrail to retain visual source coordinates and evaluate critical diagram recall needed when a handbook contains screenshots whose meaning is absent from extracted text.

B:

For prevention at the nonprofit organization, the measure to discard all images and claim diagrams are fully captured by text-only indexing is weaker or aimed at another issue than the multimodal indexing guardrail to retain visual source coordinates and evaluate critical diagram recall needed when a handbook contains screenshots whose meaning is absent from extracted text. The lasting guardrail for this specific threat is to retain visual source coordinates and evaluate critical diagram recall, not simply to complete an adjacent service task.

C:

The action to maintain evaluation for both exact and semantic intents addresses hybrid search fusion rather than the decisive multimodal indexing condition. The nonprofit organization should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to retain visual source coordinates and evaluate critical diagram recall.

Question 1025

During a service audit, a product support lead must accept a change for multimodal indexing. The issue to be addressed is: A handbook contains screenshots whose meaning is absent from extracted text. Which concrete verification action demonstrates that the relevant behavior works?

  1. Ask about a labeled screenshot component and check image-derived evidence
  2. Use supported visual extraction or image verbalization enrichment before indexing
  3. Discard all images and claim diagrams are fully captured by text-only indexing
  4. Ingest a sample and query for a semantic paraphrase

Explanation

Correct Answer

A:

The targeted acceptance test is to ask about a labeled screenshot component and check image-derived evidence. The design should first use supported visual extraction or image verbalization enrichment before indexing. A robust municipal services team deployment also needs to retain visual source coordinates and evaluate critical diagram recall; passing this check provides relevant evidence but does not certify every possible input or security risk.

Incorrect Answers

B:

The action to use supported visual extraction or image verbalization enrichment before indexing can contribute to multimodal indexing as a implementation step; here, a handbook contains screenshots whose meaning is absent from extracted text. In the municipal services team acceptance review, trying to use supported visual extraction or image verbalization enrichment before indexing does not test the exact multimodal indexing condition where a handbook contains screenshots whose meaning is absent from extracted text; the relevant test is to ask about a labeled screenshot component and check image-derived evidence.

C:

In the municipal services team acceptance review, trying to discard all images and claim diagrams are fully captured by text-only indexing does not test the exact multimodal indexing condition where a handbook contains screenshots whose meaning is absent from extracted text; the relevant test is to ask about a labeled screenshot component and check image-derived evidence. Acceptance should instead demonstrate that the service can ask about a labeled screenshot component and check image-derived evidence, using the described failure as the test criterion.

D:

In the municipal services team acceptance review, trying to ingest a sample and query for a semantic paraphrase does not test the exact multimodal indexing condition where a handbook contains screenshots whose meaning is absent from extracted text; the relevant test is to ask about a labeled screenshot component and check image-derived evidence. Without the actual change to ask about a labeled screenshot component and check image-derived evidence, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

Question 1026

At a customer support center, the data engineer is preparing a release acceptance review. The concrete requirement is: A research agent can rewrite a user query to fetch broad resources outside its scope. Which design action BEST addresses the stated need?

  1. Apply an appropriate text-splitting stage with document context retained
  2. Restrict agentic retrieval planning to authorized knowledge sources and query filters
  3. Ask a cross-tenant question and confirm retrieval remains scoped
  4. Allow generated subqueries to bypass the user's entitlement controls

Explanation

Correct Answer

B:

Choosing to restrict agentic retrieval planning to authorized knowledge sources and query filters directly targets the described agentic retrieval boundary gap: a research agent can rewrite a user query to fetch broad resources outside its scope. As a lasting boundary, they also need to test planning-generated queries for unauthorized source requests, because an implementation by itself does not prove safe operation.

Incorrect Answers

A:

Choosing to apply an appropriate text-splitting stage with document context retained cannot directly replace the specified action to restrict agentic retrieval planning to authorized knowledge sources and query filters when a research agent can rewrite a user query to fetch broad resources outside its scope. The relevant corrective or preventive action here remains to restrict agentic retrieval planning to authorized knowledge sources and query filters, with its own validation criteria.

C:

The action to ask a cross-tenant question and confirm retrieval remains scoped can contribute to agentic retrieval boundary as a targeted acceptance check step; here, a research agent can rewrite a user query to fetch broad resources outside its scope. Choosing to ask a cross-tenant question and confirm retrieval remains scoped cannot directly replace the specified action to restrict agentic retrieval planning to authorized knowledge sources and query filters when a research agent can rewrite a user query to fetch broad resources outside its scope.

D:

Choosing to allow generated subqueries to bypass the user's entitlement controls cannot directly replace the specified action to restrict agentic retrieval planning to authorized knowledge sources and query filters when a research agent can rewrite a user query to fetch broad resources outside its scope. The actual implementation must restrict agentic retrieval planning to authorized knowledge sources and query filters, because the stated service condition is that a research agent can rewrite a user query to fetch broad resources outside its scope.

Question 1027

After a design meeting at a manufacturing group, the team observes this failure: A research agent can rewrite a user query to fetch broad resources outside its scope. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Ask a cross-tenant question and confirm retrieval remains scoped
  2. Restrict agentic retrieval planning to authorized knowledge sources and query filters
  3. Expect a field that was not configured as filterable to support arbitrary filters
  4. Allow generated subqueries to bypass the user's entitlement controls

Explanation

Correct Answer

D:

The faulty assumption is allow generated subqueries to bypass the user's entitlement controls. A functioning alternative would be to restrict agentic retrieval planning to authorized knowledge sources and query filters, and the relevant validation would be to ask a cross-tenant question and confirm retrieval remains scoped.

Incorrect Answers

A:

The alternative to ask a cross-tenant question and confirm retrieval remains scoped is a targeted acceptance check step rather than the faulty assumption being diagnosed. This manufacturing group failure-analysis alternative of ask a cross-tenant question and confirm retrieval remains scoped does not diagnose the particular agentic retrieval boundary defect where a research agent can rewrite a user query to fetch broad resources outside its scope; the keyed mistake is allow generated subqueries to bypass the user's entitlement controls.

B:

The alternative to restrict agentic retrieval planning to authorized knowledge sources and query filters is a implementation step rather than the faulty assumption being diagnosed. This manufacturing group failure-analysis alternative of restrict agentic retrieval planning to authorized knowledge sources and query filters does not diagnose the particular agentic retrieval boundary defect where a research agent can rewrite a user query to fetch broad resources outside its scope; the keyed mistake is allow generated subqueries to bypass the user's entitlement controls.

C:

This manufacturing group failure-analysis alternative of expect a field that was not configured as filterable to support arbitrary filters does not diagnose the particular agentic retrieval boundary defect where a research agent can rewrite a user query to fetch broad resources outside its scope; the keyed mistake is allow generated subqueries to bypass the user's entitlement controls. The alternative may be useful in its own context, but it does not directly supply allow generated subqueries to bypass the user's entitlement controls for the stated decision.

Question 1028

The legal operations unit is responding to a staged service rollout. A cloud developer discovers this issue: A research agent can rewrite a user query to fetch broad resources outside its scope. What corrective step should the team take to restore the intended functionality most directly?

  1. Ask a cross-tenant question and confirm retrieval remains scoped
  2. Evict stale cache records and verify retrieval from the new index state
  3. Replay retrieval with enforced source access boundaries
  4. Allow generated subqueries to bypass the user's entitlement controls

Explanation

Correct Answer

C:

The appropriate remediation is to replay retrieval with enforced source access boundaries, because the failure concerns agentic retrieval boundary: a research agent can rewrite a user query to fetch broad resources outside its scope. A meaningful verification should ask a cross-tenant question and confirm retrieval remains scoped, and the durable operating rule should test planning-generated queries for unauthorized source requests after the immediate fix.

Incorrect Answers

A:

The action to ask a cross-tenant question and confirm retrieval remains scoped can contribute to agentic retrieval boundary as a targeted acceptance check step; here, a research agent can rewrite a user query to fetch broad resources outside its scope. During legal operations unit remediation, the proposed step to ask a cross-tenant question and confirm retrieval remains scoped is not the direct restoration action for agentic retrieval boundary when a research agent can rewrite a user query to fetch broad resources outside its scope; the keyed repair is replay retrieval with enforced source access boundaries.

B:

This approach relates to search cache staleness: a user sees outdated answers even though the index contains new records. If the team chose it instead, the targeted check to ask a cross-tenant question and confirm retrieval remains scoped could still fail because the underlying required action is to replay retrieval with enforced source access boundaries.

D:

During legal operations unit remediation, the proposed step to allow generated subqueries to bypass the user's entitlement controls is not the direct restoration action for agentic retrieval boundary when a research agent can rewrite a user query to fetch broad resources outside its scope; the keyed repair is replay retrieval with enforced source access boundaries. After remediation the team must ask a cross-tenant question and confirm retrieval remains scoped; the action in this option does not itself establish that target recovery outcome.

Question 1029

A site reliability engineer at a financial services division wants the following issue to be prevented in later deployments: A research agent can rewrite a user query to fetch broad resources outside its scope. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Replay retrieval with enforced source access boundaries
  2. Alert on indexing backlog quota and query saturation
  3. Allow generated subqueries to bypass the user's entitlement controls
  4. Test planning-generated queries for unauthorized source requests

Explanation

Correct Answer

D:

The sustained prevention measure is to test planning-generated queries for unauthorized source requests. The implementation can still use the separate action to restrict agentic retrieval planning to authorized knowledge sources and query filters, but a guardrail constrains future changes. The financial services division can measure whether the rule works by arranging to ask a cross-tenant question and confirm retrieval remains scoped.

Incorrect Answers

A:

The action to replay retrieval with enforced source access boundaries can contribute to agentic retrieval boundary as a corrective remediation step; here, a research agent can rewrite a user query to fetch broad resources outside its scope. For prevention at the financial services division, the measure to replay retrieval with enforced source access boundaries is weaker or aimed at another issue than the agentic retrieval boundary guardrail to test planning-generated queries for unauthorized source requests needed when a research agent can rewrite a user query to fetch broad resources outside its scope.

B:

The action to alert on indexing backlog quota and query saturation addresses rag index capacity rather than the decisive agentic retrieval boundary condition. The financial services division should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to test planning-generated queries for unauthorized source requests.

C:

For prevention at the financial services division, the measure to allow generated subqueries to bypass the user's entitlement controls is weaker or aimed at another issue than the agentic retrieval boundary guardrail to test planning-generated queries for unauthorized source requests needed when a research agent can rewrite a user query to fetch broad resources outside its scope. The lasting guardrail for this specific threat is to test planning-generated queries for unauthorized source requests, not simply to complete an adjacent service task.

Question 1030

During a design meeting, a security engineer must accept a change for agentic retrieval boundary. The issue to be addressed is: A research agent can rewrite a user query to fetch broad resources outside its scope. Which concrete verification action demonstrates that the relevant behavior works?

  1. Ask a cross-tenant question and confirm retrieval remains scoped
  2. Allow generated subqueries to bypass the user's entitlement controls
  3. Restrict agentic retrieval planning to authorized knowledge sources and query filters
  4. Ask about a labeled screenshot component and check image-derived evidence

Explanation

Correct Answer

A:

To verify this change, the logistics operator should ask a cross-tenant question and confirm retrieval remains scoped. Operators still need the implementation to restrict agentic retrieval planning to authorized knowledge sources and query filters and the preventive requirement to test planning-generated queries for unauthorized source requests.

Incorrect Answers

B:

In the logistics operator acceptance review, trying to allow generated subqueries to bypass the user's entitlement controls does not test the exact agentic retrieval boundary condition where a research agent can rewrite a user query to fetch broad resources outside its scope; the relevant test is to ask a cross-tenant question and confirm retrieval remains scoped. Acceptance should instead demonstrate that the service can ask a cross-tenant question and confirm retrieval remains scoped, using the described failure as the test criterion.

C:

The action to restrict agentic retrieval planning to authorized knowledge sources and query filters can contribute to agentic retrieval boundary as a implementation step; here, a research agent can rewrite a user query to fetch broad resources outside its scope. In the logistics operator acceptance review, trying to restrict agentic retrieval planning to authorized knowledge sources and query filters does not test the exact agentic retrieval boundary condition where a research agent can rewrite a user query to fetch broad resources outside its scope; the relevant test is to ask a cross-tenant question and confirm retrieval remains scoped.

D:

For multimodal indexing, it can be sensible to ask about a labeled screenshot component and check image-derived evidence if a handbook contains screenshots whose meaning is absent from extracted text. Without the actual change to ask a cross-tenant question and confirm retrieval remains scoped, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

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