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 2

Question 10

During a production incident, a responsible-AI reviewer must accept a change for multimodal support selection. The issue to be addressed is: An intake app must reason jointly about photographs and textual descriptions. Which concrete verification action demonstrates that the relevant behavior works?

  1. Simulate two tenant sessions and verify no preference crosses the boundary
  2. Submit paired images and descriptions and check image-grounded responses
  3. Select a multimodal-capable model that accepts both image and text inputs
  4. Treat a text-only model as though it can inspect image pixels

Explanation

Correct Answer

B:

To verify this change, the payment services team should submit paired images and descriptions and check image-grounded responses. Operators still need the implementation to select a multimodal-capable model that accepts both image and text inputs and the preventive requirement to validate supported modalities and input limits before rollout.

Incorrect Answers

A:

For agent memory scope, it can be sensible to simulate two tenant sessions and verify no preference crosses the boundary if an assistant must recall preferences within a permitted client session but isolate different clients. Without the actual change to submit paired images and descriptions and check image-grounded responses, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

C:

The action to select a multimodal-capable model that accepts both image and text inputs can contribute to multimodal support selection as a implementation step; here, an intake app must reason jointly about photographs and textual descriptions. In the payment services team acceptance review, trying to select a multimodal-capable model that accepts both image and text inputs does not test the exact multimodal support selection condition where an intake app must reason jointly about photographs and textual descriptions; the relevant test is to submit paired images and descriptions and check image-grounded responses.

D:

In the payment services team acceptance review, trying to treat a text-only model as though it can inspect image pixels does not test the exact multimodal support selection condition where an intake app must reason jointly about photographs and textual descriptions; the relevant test is to submit paired images and descriptions and check image-grounded responses. Acceptance should instead demonstrate that the service can submit paired images and descriptions and check image-grounded responses, using the described failure as the test criterion.

Question 11

At a manufacturing group, the solution architect is preparing a architecture migration. The concrete requirement is: A document assistant truncates key context when submitting long cases. Which design action BEST addresses the stated need?

  1. Send a long case with decisive evidence near its end and inspect the answer
  2. Evaluate supported Foundry catalog models against modality quality cost and available region
  3. Monitor token budgets and require tests near documented limits
  4. Choose a supported context window and retrieval strategy that fits the relevant evidence
  5. Assume all advertised model families accept identical context limits

Explanation

Correct Answer

D:

The proposed implementation is to choose a supported context window and retrieval strategy that fits the relevant evidence; this matches the operational requirement that a document assistant truncates key context when submitting long cases. The team can send a long case with decisive evidence near its end and inspect the answer. The team should also monitor token budgets and require tests near documented limits.

Incorrect Answers

A:

The action to send a long case with decisive evidence near its end and inspect the answer can contribute to generative model context size as a targeted acceptance check step; here, a document assistant truncates key context when submitting long cases. Choosing to send a long case with decisive evidence near its end and inspect the answer cannot directly replace the specified action to choose a supported context window and retrieval strategy that fits the relevant evidence when a document assistant truncates key context when submitting long cases.

B:

The option to evaluate supported Foundry catalog models against modality quality cost and available region would normally serve a different foundry model catalog selection requirement, where a workload must choose among available models with distinct capabilities and regional support. The alternative may be useful in its own context, but it does not directly supply choose a supported context window and retrieval strategy that fits the relevant evidence for the stated decision.

C:

The action to monitor token budgets and require tests near documented limits can contribute to generative model context size as a ongoing preventive rule step; here, a document assistant truncates key context when submitting long cases. Choosing to monitor token budgets and require tests near documented limits cannot directly replace the specified action to choose a supported context window and retrieval strategy that fits the relevant evidence when a document assistant truncates key context when submitting long cases.

E:

Choosing to assume all advertised model families accept identical context limits cannot directly replace the specified action to choose a supported context window and retrieval strategy that fits the relevant evidence when a document assistant truncates key context when submitting long cases. The actual implementation must choose a supported context window and retrieval strategy that fits the relevant evidence, because the stated service condition is that a document assistant truncates key context when submitting long cases.

Question 12

After a operational readiness review at a legal operations unit, the team observes this failure: A document assistant truncates key context when submitting long cases. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Route to an untested model without checking tool or modality support
  2. Choose a supported context window and retrieval strategy that fits the relevant evidence
  3. Send a long case with decisive evidence near its end and inspect the answer
  4. Assume all advertised model families accept identical context limits

Explanation

Correct Answer

D:

This design mistake is to assume all advertised model families accept identical context limits; it conflicts with the actual generative model context size behavior needed when a document assistant truncates key context when submitting long cases. The legal operations unit should instead choose a supported context window and retrieval strategy that fits the relevant evidence, followed by a check to send a long case with decisive evidence near its end and inspect the answer.

Incorrect Answers

A:

This legal operations unit failure-analysis alternative of route to an untested model without checking tool or modality support does not diagnose the particular generative model context size defect where a document assistant truncates key context when submitting long cases; the keyed mistake is assume all advertised model families accept identical context limits. The alternative may be useful in its own context, but it does not directly supply assume all advertised model families accept identical context limits for the stated decision.

B:

The alternative to choose a supported context window and retrieval strategy that fits the relevant evidence is a implementation step rather than the faulty assumption being diagnosed. This legal operations unit failure-analysis alternative of choose a supported context window and retrieval strategy that fits the relevant evidence does not diagnose the particular generative model context size defect where a document assistant truncates key context when submitting long cases; the keyed mistake is assume all advertised model families accept identical context limits.

C:

The alternative to send a long case with decisive evidence near its end and inspect the answer is a targeted acceptance check step rather than the faulty assumption being diagnosed. This legal operations unit failure-analysis alternative of send a long case with decisive evidence near its end and inspect the answer does not diagnose the particular generative model context size defect where a document assistant truncates key context when submitting long cases; the keyed mistake is assume all advertised model families accept identical context limits.

Question 13

The financial services division is responding to a regional deployment. A AI application owner discovers this issue: A document assistant truncates key context when submitting long cases. What corrective step should the team take to restore the intended functionality most directly?

  1. Assume all advertised model families accept identical context limits
  2. Send a long case with decisive evidence near its end and inspect the answer
  3. Separate memory stores by tenant and revoke cross-tenant cache keys
  4. Trim irrelevant retrieval and select a verified context configuration

Explanation

Correct Answer

D:

The direct repair is to trim irrelevant retrieval and select a verified context configuration. Once the condition is corrected, the financial services division should send a long case with decisive evidence near its end and inspect the answer. To reduce recurrence, it should also monitor token budgets and require tests near documented limits; those steps provide different recovery and prevention functions.

Incorrect Answers

A:

During financial services division remediation, the proposed step to assume all advertised model families accept identical context limits is not the direct restoration action for generative model context size when a document assistant truncates key context when submitting long cases; the keyed repair is trim irrelevant retrieval and select a verified context configuration. After remediation the team must send a long case with decisive evidence near its end and inspect the answer; the action in this option does not itself establish that target recovery outcome.

B:

The action to send a long case with decisive evidence near its end and inspect the answer can contribute to generative model context size as a targeted acceptance check step; here, a document assistant truncates key context when submitting long cases. During financial services division remediation, the proposed step to send a long case with decisive evidence near its end and inspect the answer is not the direct restoration action for generative model context size when a document assistant truncates key context when submitting long cases; the keyed repair is trim irrelevant retrieval and select a verified context configuration.

C:

This approach relates to agent memory scope: an assistant must recall preferences within a permitted client session but isolate different clients. If the team chose it instead, the targeted check to send a long case with decisive evidence near its end and inspect the answer could still fail because the underlying required action is to trim irrelevant retrieval and select a verified context configuration.

Question 14

A quality assurance lead at a logistics operator wants the following issue to be prevented in later deployments: A document assistant truncates key context when submitting long cases. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Assume all advertised model families accept identical context limits
  2. Restrict tool permissions and validate arguments before every call
  3. Trim irrelevant retrieval and select a verified context configuration
  4. Monitor token budgets and require tests near documented limits

Explanation

Correct Answer

D:

Standardizing the boundary to monitor token budgets and require tests near documented limits is preventive because the same generative model context size failure could otherwise return: a document assistant truncates key context when submitting long cases. For this logistics operator, the service should also choose a supported context window and retrieval strategy that fits the relevant evidence, and tests should send a long case with decisive evidence near its end and inspect the answer so that the ongoing requirement is measurable.

Incorrect Answers

A:

For prevention at the logistics operator, the measure to assume all advertised model families accept identical context limits is weaker or aimed at another issue than the generative model context size guardrail to monitor token budgets and require tests near documented limits needed when a document assistant truncates key context when submitting long cases. The lasting guardrail for this specific threat is to monitor token budgets and require tests near documented limits, not simply to complete an adjacent service task.

B:

The action to restrict tool permissions and validate arguments before every call addresses tool calling versus freeform text rather than the decisive generative model context size condition. The logistics operator should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to monitor token budgets and require tests near documented limits.

C:

The action to trim irrelevant retrieval and select a verified context configuration can contribute to generative model context size as a corrective remediation step; here, a document assistant truncates key context when submitting long cases. For prevention at the logistics operator, the measure to trim irrelevant retrieval and select a verified context configuration is weaker or aimed at another issue than the generative model context size guardrail to monitor token budgets and require tests near documented limits needed when a document assistant truncates key context when submitting long cases.

Question 15

During a operational readiness review, a AI platform engineer must accept a change for generative model context size. The issue to be addressed is: A document assistant truncates key context when submitting long cases. Which concrete verification action demonstrates that the relevant behavior works?

  1. Assume all advertised model families accept identical context limits
  2. Choose a supported context window and retrieval strategy that fits the relevant evidence
  3. Send a long case with decisive evidence near its end and inspect the answer
  4. Force the primary endpoint offline and confirm safe fallback behavior

Explanation

Correct Answer

C:

The targeted acceptance test is to send a long case with decisive evidence near its end and inspect the answer. The design should first choose a supported context window and retrieval strategy that fits the relevant evidence. A robust energy cooperative deployment also needs to monitor token budgets and require tests near documented limits; passing this check provides relevant evidence but does not certify every possible input or security risk.

Incorrect Answers

A:

In the energy cooperative acceptance review, trying to assume all advertised model families accept identical context limits does not test the exact generative model context size condition where a document assistant truncates key context when submitting long cases; the relevant test is to send a long case with decisive evidence near its end and inspect the answer. Acceptance should instead demonstrate that the service can send a long case with decisive evidence near its end and inspect the answer, using the described failure as the test criterion.

B:

The action to choose a supported context window and retrieval strategy that fits the relevant evidence can contribute to generative model context size as a implementation step; here, a document assistant truncates key context when submitting long cases. In the energy cooperative acceptance review, trying to choose a supported context window and retrieval strategy that fits the relevant evidence does not test the exact generative model context size condition where a document assistant truncates key context when submitting long cases; the relevant test is to send a long case with decisive evidence near its end and inspect the answer.

D:

For model fallback planning, it can be sensible to force the primary endpoint offline and confirm safe fallback behavior if a critical assistant needs continuity when the preferred deployment is unavailable. Without the actual change to send a long case with decisive evidence near its end and inspect the answer, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

Question 16

At a payment services team, the compliance specialist is preparing a cost review. The concrete requirement is: A search service must match paraphrases across internal manuals. Which design action BEST addresses the stated need?

  1. Compare raw text strings alone while promising semantic vector retrieval
  2. Select a tested alternate compatible model and preserve policy and quality safeguards
  3. Create document and query embeddings using compatible embedding configurations
  4. Query paraphrases and verify relevant chunks rank in returned results

Explanation

Correct Answer

C:

Choosing to create document and query embeddings using compatible embedding configurations directly targets the described embeddings for semantic retrieval gap: a search service must match paraphrases across internal manuals. As a lasting boundary, they also need to version embedding deployments and reindex after incompatible changes, because an implementation by itself does not prove safe operation.

Incorrect Answers

A:

Choosing to compare raw text strings alone while promising semantic vector retrieval cannot directly replace the specified action to create document and query embeddings using compatible embedding configurations when a search service must match paraphrases across internal manuals. The actual implementation must create document and query embeddings using compatible embedding configurations, because the stated service condition is that a search service must match paraphrases across internal manuals.

B:

Choosing to select a tested alternate compatible model and preserve policy and quality safeguards would direct effort toward model fallback planning, whose characteristic issue is that a critical assistant needs continuity when the preferred deployment is unavailable. The relevant corrective or preventive action here remains to create document and query embeddings using compatible embedding configurations, with its own validation criteria.

D:

The action to query paraphrases and verify relevant chunks rank in returned results can contribute to embeddings for semantic retrieval as a targeted acceptance check step; here, a search service must match paraphrases across internal manuals. Choosing to query paraphrases and verify relevant chunks rank in returned results cannot directly replace the specified action to create document and query embeddings using compatible embedding configurations when a search service must match paraphrases across internal manuals.

Question 17

After a reliability exercise at a transport authority, the team observes this failure: A search service must match paraphrases across internal manuals. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Create document and query embeddings using compatible embedding configurations
  2. Treat a model saying it called a tool as proof an API was executed
  3. Compare raw text strings alone while promising semantic vector retrieval
  4. Query paraphrases and verify relevant chunks rank in returned results
  5. Version embedding deployments and reindex after incompatible changes
  6. Treat an arbitrary single still image as a complete description of motion

Explanation

Correct Answer

C:

The faulty assumption is compare raw text strings alone while promising semantic vector retrieval. A functioning alternative would be to create document and query embeddings using compatible embedding configurations, and the relevant validation would be to query paraphrases and verify relevant chunks rank in returned results.

Incorrect Answers

A:

The alternative to create document and query embeddings using compatible embedding configurations is a implementation step rather than the faulty assumption being diagnosed. This transport authority failure-analysis alternative of create document and query embeddings using compatible embedding configurations does not diagnose the particular embeddings for semantic retrieval defect where a search service must match paraphrases across internal manuals; the keyed mistake is compare raw text strings alone while promising semantic vector retrieval.

B:

The action to treat a model saying it called a tool as proof an API was executed addresses tool calling versus freeform text rather than the decisive embeddings for semantic retrieval condition. The transport authority should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to compare raw text strings alone while promising semantic vector retrieval.

D:

The alternative to query paraphrases and verify relevant chunks rank in returned results is a targeted acceptance check step rather than the faulty assumption being diagnosed. This transport authority failure-analysis alternative of query paraphrases and verify relevant chunks rank in returned results does not diagnose the particular embeddings for semantic retrieval defect where a search service must match paraphrases across internal manuals; the keyed mistake is compare raw text strings alone while promising semantic vector retrieval.

E:

The alternative to version embedding deployments and reindex after incompatible changes is a ongoing preventive rule step rather than the faulty assumption being diagnosed. This transport authority failure-analysis alternative of version embedding deployments and reindex after incompatible changes does not diagnose the particular embeddings for semantic retrieval defect where a search service must match paraphrases across internal manuals; the keyed mistake is compare raw text strings alone while promising semantic vector retrieval.

F:

This approach relates to video context selection: a safety application must summarize events occurring across video frames. If the team chose it instead, the targeted check to query paraphrases and verify relevant chunks rank in returned results could still fail because the underlying required action is to compare raw text strings alone while promising semantic vector retrieval.

Question 18

The large online retailer is responding to a source update. A data engineer discovers this issue: A search service must match paraphrases across internal manuals. What corrective step should the team take to restore the intended functionality most directly?

  1. Re-embed mismatched documents and queries under a consistent representation
  2. Switch to a supported standby deployment and rerun regression prompts
  3. Compare raw text strings alone while promising semantic vector retrieval
  4. Query paraphrases and verify relevant chunks rank in returned results

Explanation

Correct Answer

A:

The appropriate remediation is to re-embed mismatched documents and queries under a consistent representation, because the failure concerns embeddings for semantic retrieval: a search service must match paraphrases across internal manuals. A meaningful verification should query paraphrases and verify relevant chunks rank in returned results, and the durable operating rule should version embedding deployments and reindex after incompatible changes after the immediate fix.

Incorrect Answers

B:

This approach relates to model fallback planning: a critical assistant needs continuity when the preferred deployment is unavailable. If the team chose it instead, the targeted check to query paraphrases and verify relevant chunks rank in returned results could still fail because the underlying required action is to re-embed mismatched documents and queries under a consistent representation.

C:

During large online retailer remediation, the proposed step to compare raw text strings alone while promising semantic vector retrieval is not the direct restoration action for embeddings for semantic retrieval when a search service must match paraphrases across internal manuals; the keyed repair is re-embed mismatched documents and queries under a consistent representation. After remediation the team must query paraphrases and verify relevant chunks rank in returned results; the action in this option does not itself establish that target recovery outcome.

D:

The action to query paraphrases and verify relevant chunks rank in returned results can contribute to embeddings for semantic retrieval as a targeted acceptance check step; here, a search service must match paraphrases across internal manuals. During large online retailer remediation, the proposed step to query paraphrases and verify relevant chunks rank in returned results is not the direct restoration action for embeddings for semantic retrieval when a search service must match paraphrases across internal manuals; the keyed repair is re-embed mismatched documents and queries under a consistent representation.

Question 19

A service administrator at a public library network wants the following issue to be prevented in later deployments: A search service must match paraphrases across internal manuals. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Enforce accuracy and latency acceptance thresholds in model selection
  2. Re-embed mismatched documents and queries under a consistent representation
  3. Compare raw text strings alone while promising semantic vector retrieval
  4. Version embedding deployments and reindex after incompatible changes

Explanation

Correct Answer

D:

The sustained prevention measure is to version embedding deployments and reindex after incompatible changes. The implementation can still use the separate action to create document and query embeddings using compatible embedding configurations, but a guardrail constrains future changes. The public library network can measure whether the rule works by arranging to query paraphrases and verify relevant chunks rank in returned results.

Incorrect Answers

A:

The action to enforce accuracy and latency acceptance thresholds in model selection addresses small-model latency tradeoff rather than the decisive embeddings for semantic retrieval condition. The public library network should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to version embedding deployments and reindex after incompatible changes.

B:

The action to re-embed mismatched documents and queries under a consistent representation can contribute to embeddings for semantic retrieval as a corrective remediation step; here, a search service must match paraphrases across internal manuals. For prevention at the public library network, the measure to re-embed mismatched documents and queries under a consistent representation is weaker or aimed at another issue than the embeddings for semantic retrieval guardrail to version embedding deployments and reindex after incompatible changes needed when a search service must match paraphrases across internal manuals.

C:

For prevention at the public library network, the measure to compare raw text strings alone while promising semantic vector retrieval is weaker or aimed at another issue than the embeddings for semantic retrieval guardrail to version embedding deployments and reindex after incompatible changes needed when a search service must match paraphrases across internal manuals. The lasting guardrail for this specific threat is to version embedding deployments and reindex after incompatible changes, not simply to complete an adjacent service task.

Question 20

During a reliability exercise, a cloud developer must accept a change for embeddings for semantic retrieval. The issue to be addressed is: A search service must match paraphrases across internal manuals. Which concrete verification action demonstrates that the relevant behavior works?

  1. Create document and query embeddings using compatible embedding configurations
  2. Compare raw text strings alone while promising semantic vector retrieval
  3. Query paraphrases and verify relevant chunks rank in returned results
  4. Run representative spoken queries and measure transcription and playback

Explanation

Correct Answer

C:

To verify this change, the software consultancy should query paraphrases and verify relevant chunks rank in returned results. Operators still need the implementation to create document and query embeddings using compatible embedding configurations and the preventive requirement to version embedding deployments and reindex after incompatible changes.

Incorrect Answers

A:

The action to create document and query embeddings using compatible embedding configurations can contribute to embeddings for semantic retrieval as a implementation step; here, a search service must match paraphrases across internal manuals. In the software consultancy acceptance review, trying to create document and query embeddings using compatible embedding configurations does not test the exact embeddings for semantic retrieval condition where a search service must match paraphrases across internal manuals; the relevant test is to query paraphrases and verify relevant chunks rank in returned results.

B:

In the software consultancy acceptance review, trying to compare raw text strings alone while promising semantic vector retrieval does not test the exact embeddings for semantic retrieval condition where a search service must match paraphrases across internal manuals; the relevant test is to query paraphrases and verify relevant chunks rank in returned results. Acceptance should instead demonstrate that the service can query paraphrases and verify relevant chunks rank in returned results, using the described failure as the test criterion.

D:

For speech modality selection, it can be sensible to run representative spoken queries and measure transcription and playback if a voice support assistant must handle live speech and produce spoken answers. Without the actual change to query paraphrases and verify relevant chunks rank in returned results, 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