Microsoft AI-103 Dumps
Microsoft AI-103 Practice Test 35
Question 340
During a architecture migration, a data engineer must accept a change for hybrid retrieval for identifiers. The issue to be addressed is: Users search for part numbers and semantically related troubleshooting symptoms. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
To verify this change, the payment services team should retrieve an exact part number and a paraphrased fault description. Operators still need the implementation to combine exact keyword and vector retrieval and evaluate ranking quality and the preventive requirement to maintain dual query tests for identifier precision and paraphrase recall.
Incorrect Answers
The action to add lexical matching and rerank merged relevant results can contribute to hybrid retrieval for identifiers as a corrective remediation step; here, users search for part numbers and semantically related troubleshooting symptoms. In the payment services team acceptance review, trying to add lexical matching and rerank merged relevant results does not test the exact hybrid retrieval for identifiers condition where users search for part numbers and semantically related troubleshooting symptoms; the relevant test is to retrieve an exact part number and a paraphrased fault description.
The action to maintain dual query tests for identifier precision and paraphrase recall can contribute to hybrid retrieval for identifiers as a ongoing preventive rule step; here, users search for part numbers and semantically related troubleshooting symptoms. In the payment services team acceptance review, trying to maintain dual query tests for identifier precision and paraphrase recall does not test the exact hybrid retrieval for identifiers condition where users search for part numbers and semantically related troubleshooting symptoms; the relevant test is to retrieve an exact part number and a paraphrased fault description.
The action to combine exact keyword and vector retrieval and evaluate ranking quality can contribute to hybrid retrieval for identifiers as a implementation step; here, users search for part numbers and semantically related troubleshooting symptoms. In the payment services team acceptance review, trying to combine exact keyword and vector retrieval and evaluate ranking quality does not test the exact hybrid retrieval for identifiers condition where users search for part numbers and semantically related troubleshooting symptoms; the relevant test is to retrieve an exact part number and a paraphrased fault description.
The option to repeat the same labeled inputs and compare output variability would normally serve a different temperature for stable extraction requirement, where a classification service must return reproducible labels with limited creative variation. The alternative may be useful in its own context, but it does not directly supply retrieve an exact part number and a paraphrased fault description for the stated decision.
In the payment services team acceptance review, trying to depend only on semantic vectors for exact serial numbers without testing recall does not test the exact hybrid retrieval for identifiers condition where users search for part numbers and semantically related troubleshooting symptoms; the relevant test is to retrieve an exact part number and a paraphrased fault description. Acceptance should instead demonstrate that the service can retrieve an exact part number and a paraphrased fault description, using the described failure as the test criterion.
Question 341
A service administrator at a manufacturing group is conducting a source update concerning semantic reranking. The observed problem is: Search results contain relevant snippets far below unrelated vector matches. Which TWO actions together represent the needed implementation AND a durable preventive safeguard? (Select TWO answers.)
Explanation
Correct Answers
The proposed implementation is to apply supported semantic ranking on retrieved candidates with suitable text fields; this matches the operational requirement that search results contain relevant snippets far below unrelated vector matches. The team can compare labeled query relevance before and after semantic reranking. The team should also evaluate first-stage recall separately from reranker usefulness.
The sustained prevention measure is to evaluate first-stage recall separately from reranker usefulness. The implementation can still use the separate action to apply supported semantic ranking on retrieved candidates with suitable text fields, but a guardrail constrains future changes. The manufacturing group can measure whether the rule works by arranging to compare labeled query relevance before and after semantic reranking.
Incorrect Answers
The option to use layout-aware parsing and explicit column-row associations for table extraction would normally serve a different grounded table question answering requirement, where a sales report assistant confuses figures across table headers. The alternative may be useful in its own context, but it does not directly supply apply supported semantic ranking on retrieved candidates with suitable text fields for the stated decision.
Choosing to assume semantic reranking can fix an empty or incorrectly filtered candidate set cannot directly replace the specified action to apply supported semantic ranking on retrieved candidates with suitable text fields when search results contain relevant snippets far below unrelated vector matches. The actual implementation must apply supported semantic ranking on retrieved candidates with suitable text fields, because the stated service condition is that search results contain relevant snippets far below unrelated vector matches.
The action to compare labeled query relevance before and after semantic reranking can contribute to semantic reranking as a targeted acceptance check step; here, search results contain relevant snippets far below unrelated vector matches. Choosing to compare labeled query relevance before and after semantic reranking cannot directly replace the specified action to apply supported semantic ranking on retrieved candidates with suitable text fields when search results contain relevant snippets far below unrelated vector matches.
Question 342
After a cost review at a legal operations unit, the team observes this failure: Search results contain relevant snippets far below unrelated vector matches. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?
Explanation
Correct Answer
This design mistake is to assume semantic reranking can fix an empty or incorrectly filtered candidate set; it conflicts with the actual semantic reranking behavior needed when search results contain relevant snippets far below unrelated vector matches. The legal operations unit should instead apply supported semantic ranking on retrieved candidates with suitable text fields, followed by a check to compare labeled query relevance before and after semantic reranking.
Incorrect Answers
The alternative to apply supported semantic ranking on retrieved candidates with suitable text fields is a implementation step rather than the faulty assumption being diagnosed. This legal operations unit failure-analysis alternative of apply supported semantic ranking on retrieved candidates with suitable text fields does not diagnose the particular semantic reranking defect where search results contain relevant snippets far below unrelated vector matches; the keyed mistake is assume semantic reranking can fix an empty or incorrectly filtered candidate set.
The alternative to compare labeled query relevance before and after semantic reranking is a targeted acceptance check step rather than the faulty assumption being diagnosed. This legal operations unit failure-analysis alternative of compare labeled query relevance before and after semantic reranking does not diagnose the particular semantic reranking defect where search results contain relevant snippets far below unrelated vector matches; the keyed mistake is assume semantic reranking can fix an empty or incorrectly filtered candidate set.
The option to increase generation randomness to guarantee exact repeated labels would normally serve a different temperature for stable extraction requirement, where a classification service must return reproducible labels with limited creative variation. The alternative may be useful in its own context, but it does not directly supply assume semantic reranking can fix an empty or incorrectly filtered candidate set for the stated decision.
Question 343
The financial services division is responding to a reliability exercise. A site reliability engineer discovers this issue: Search results contain relevant snippets far below unrelated vector matches. What corrective step should the team take to restore the intended functionality most directly?
Explanation
Correct Answer
The direct repair is to repair candidate retrieval and configure reranking for good fields. Once the condition is corrected, the financial services division should compare labeled query relevance before and after semantic reranking. To reduce recurrence, it should also evaluate first-stage recall separately from reranker usefulness; those steps provide different recovery and prevention functions.
Incorrect Answers
The action to compare labeled query relevance before and after semantic reranking can contribute to semantic reranking as a targeted acceptance check step; here, search results contain relevant snippets far below unrelated vector matches. During financial services division remediation, the proposed step to compare labeled query relevance before and after semantic reranking is not the direct restoration action for semantic reranking when search results contain relevant snippets far below unrelated vector matches; the keyed repair is repair candidate retrieval and configure reranking for good fields.
This approach relates to sampling for creative drafting: a marketing draft assistant produces nearly identical copy despite requests for alternatives. If the team chose it instead, the targeted check to compare labeled query relevance before and after semantic reranking could still fail because the underlying required action is to repair candidate retrieval and configure reranking for good fields.
During financial services division remediation, the proposed step to assume semantic reranking can fix an empty or incorrectly filtered candidate set is not the direct restoration action for semantic reranking when search results contain relevant snippets far below unrelated vector matches; the keyed repair is repair candidate retrieval and configure reranking for good fields. After remediation the team must compare labeled query relevance before and after semantic reranking; the action in this option does not itself establish that target recovery outcome.
Question 344
A security engineer at a logistics operator wants the following issue to be prevented in later deployments: Search results contain relevant snippets far below unrelated vector matches. Which enduring control or engineering guardrail BEST prevents recurrence?
Explanation
Correct Answer
Standardizing the boundary to evaluate first-stage recall separately from reranker usefulness is preventive because the same semantic reranking failure could otherwise return: search results contain relevant snippets far below unrelated vector matches. For this logistics operator, the service should also apply supported semantic ranking on retrieved candidates with suitable text fields, and tests should compare labeled query relevance before and after semantic reranking so that the ongoing requirement is measurable.
Incorrect Answers
For prevention at the logistics operator, the measure to assume semantic reranking can fix an empty or incorrectly filtered candidate set is weaker or aimed at another issue than the semantic reranking guardrail to evaluate first-stage recall separately from reranker usefulness needed when search results contain relevant snippets far below unrelated vector matches. The lasting guardrail for this specific threat is to evaluate first-stage recall separately from reranker usefulness, not simply to complete an adjacent service task.
The action to maintain versioned factual evaluation examples linked to original sources addresses model evaluation for fabrication rather than the decisive semantic reranking condition. The logistics operator should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to evaluate first-stage recall separately from reranker usefulness.
The action to repair candidate retrieval and configure reranking for good fields can contribute to semantic reranking as a corrective remediation step; here, search results contain relevant snippets far below unrelated vector matches. For prevention at the logistics operator, the measure to repair candidate retrieval and configure reranking for good fields is weaker or aimed at another issue than the semantic reranking guardrail to evaluate first-stage recall separately from reranker usefulness needed when search results contain relevant snippets far below unrelated vector matches.
Question 345
During a cost review, a agent designer must accept a change for semantic reranking. The issue to be addressed is: Search results contain relevant snippets far below unrelated vector matches. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
The targeted acceptance test is to compare labeled query relevance before and after semantic reranking. The design should first apply supported semantic ranking on retrieved candidates with suitable text fields. A robust energy cooperative deployment also needs to evaluate first-stage recall separately from reranker usefulness; passing this check provides relevant evidence but does not certify every possible input or security risk.
Incorrect Answers
For citation alignment, it can be sensible to manually inspect citation spans on a difficult evaluation set if an assistant cites a policy file unrelated to the answer it provides. Without the actual change to compare labeled query relevance before and after semantic reranking, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.
In the energy cooperative acceptance review, trying to assume semantic reranking can fix an empty or incorrectly filtered candidate set does not test the exact semantic reranking condition where search results contain relevant snippets far below unrelated vector matches; the relevant test is to compare labeled query relevance before and after semantic reranking. Acceptance should instead demonstrate that the service can compare labeled query relevance before and after semantic reranking, using the described failure as the test criterion.
The action to apply supported semantic ranking on retrieved candidates with suitable text fields can contribute to semantic reranking as a implementation step; here, search results contain relevant snippets far below unrelated vector matches. In the energy cooperative acceptance review, trying to apply supported semantic ranking on retrieved candidates with suitable text fields does not test the exact semantic reranking condition where search results contain relevant snippets far below unrelated vector matches; the relevant test is to compare labeled query relevance before and after semantic reranking.
Question 346
At a payment services team, the responsible-AI reviewer is preparing a support escalation. The concrete requirement is: An assistant cites a policy file unrelated to the answer it provides. Which design action BEST addresses the stated need?
Explanation
Correct Answer
Choosing to map generated claims to specific retrieved evidence spans and reject unsupported claims directly targets the described citation alignment gap: an assistant cites a policy file unrelated to the answer it provides. As a lasting boundary, they also need to validate citation-to-claim evidence and source version, because an implementation by itself does not prove safe operation.
Incorrect Answers
The action to manually inspect citation spans on a difficult evaluation set can contribute to citation alignment as a targeted acceptance check step; here, an assistant cites a policy file unrelated to the answer it provides. Choosing to manually inspect citation spans on a difficult evaluation set cannot directly replace the specified action to map generated claims to specific retrieved evidence spans and reject unsupported claims when an assistant cites a policy file unrelated to the answer it provides.
Choosing to attach a random retrieved URL and call every answer source-grounded cannot directly replace the specified action to map generated claims to specific retrieved evidence spans and reject unsupported claims when an assistant cites a policy file unrelated to the answer it provides. The actual implementation must map generated claims to specific retrieved evidence spans and reject unsupported claims, because the stated service condition is that an assistant cites a policy file unrelated to the answer it provides.
Choosing to apply supported semantic ranking on retrieved candidates with suitable text fields would direct effort toward semantic reranking, whose characteristic issue is that search results contain relevant snippets far below unrelated vector matches. The relevant corrective or preventive action here remains to map generated claims to specific retrieved evidence spans and reject unsupported claims, with its own validation criteria.
Question 347
After a service audit at a transport authority, the team observes this failure: An assistant cites a policy file unrelated to the answer it provides. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?
Explanation
Correct Answer
The faulty assumption is attach a random retrieved URL and call every answer source-grounded. A functioning alternative would be to map generated claims to specific retrieved evidence spans and reject unsupported claims, and the relevant validation would be to manually inspect citation spans on a difficult evaluation set.
Incorrect Answers
The alternative to manually inspect citation spans on a difficult evaluation set is a targeted acceptance check step rather than the faulty assumption being diagnosed. This transport authority failure-analysis alternative of manually inspect citation spans on a difficult evaluation set does not diagnose the particular citation alignment defect where an assistant cites a policy file unrelated to the answer it provides; the keyed mistake is attach a random retrieved URL and call every answer source-grounded.
The option to assume prompting for JSON guarantees syntactically valid and typed data would normally serve a different structured json generation requirement, where a claims workflow requires parseable fields rather than an unconstrained narrative. The alternative may be useful in its own context, but it does not directly supply attach a random retrieved URL and call every answer source-grounded for the stated decision.
The alternative to map generated claims to specific retrieved evidence spans and reject unsupported claims is a implementation step rather than the faulty assumption being diagnosed. This transport authority failure-analysis alternative of map generated claims to specific retrieved evidence spans and reject unsupported claims does not diagnose the particular citation alignment defect where an assistant cites a policy file unrelated to the answer it provides; the keyed mistake is attach a random retrieved URL and call every answer source-grounded.
Question 348
The large online retailer is responding to a security tabletop exercise. A machine-learning developer discovers this issue: An assistant cites a policy file unrelated to the answer it provides. What corrective step should the team take to restore the intended functionality most directly?
Explanation
Correct Answer
The appropriate remediation is to regenerate the answer using the correct supported passages, because the failure concerns citation alignment: an assistant cites a policy file unrelated to the answer it provides. A meaningful verification should manually inspect citation spans on a difficult evaluation set, and the durable operating rule should validate citation-to-claim evidence and source version after the immediate fix.
Incorrect Answers
This approach relates to retrieval-augmented policy answers: a benefits assistant must use current private HR policies. If the team chose it instead, the targeted check to manually inspect citation spans on a difficult evaluation set could still fail because the underlying required action is to regenerate the answer using the correct supported passages.
The action to manually inspect citation spans on a difficult evaluation set can contribute to citation alignment as a targeted acceptance check step; here, an assistant cites a policy file unrelated to the answer it provides. During large online retailer remediation, the proposed step to manually inspect citation spans on a difficult evaluation set is not the direct restoration action for citation alignment when an assistant cites a policy file unrelated to the answer it provides; the keyed repair is regenerate the answer using the correct supported passages.
During large online retailer remediation, the proposed step to attach a random retrieved URL and call every answer source-grounded is not the direct restoration action for citation alignment when an assistant cites a policy file unrelated to the answer it provides; the keyed repair is regenerate the answer using the correct supported passages. After remediation the team must manually inspect citation spans on a difficult evaluation set; the action in this option does not itself establish that target recovery outcome.
Question 349
A AI application owner at a public library network wants the following issue to be prevented in later deployments: An assistant cites a policy file unrelated to the answer it provides. Which enduring control or engineering guardrail BEST prevents recurrence?
Explanation
Correct Answer
The sustained prevention measure is to validate citation-to-claim evidence and source version. The implementation can still use the separate action to map generated claims to specific retrieved evidence spans and reject unsupported claims, but a guardrail constrains future changes. The public library network can measure whether the rule works by arranging to manually inspect citation spans on a difficult evaluation set.
Incorrect Answers
The action to maintain versioned factual evaluation examples linked to original sources addresses model evaluation for fabrication rather than the decisive citation alignment condition. The public library network should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to validate citation-to-claim evidence and source version.
The action to regenerate the answer using the correct supported passages can contribute to citation alignment as a corrective remediation step; here, an assistant cites a policy file unrelated to the answer it provides. For prevention at the public library network, the measure to regenerate the answer using the correct supported passages is weaker or aimed at another issue than the citation alignment guardrail to validate citation-to-claim evidence and source version needed when an assistant cites a policy file unrelated to the answer it provides.
For prevention at the public library network, the measure to attach a random retrieved URL and call every answer source-grounded is weaker or aimed at another issue than the citation alignment guardrail to validate citation-to-claim evidence and source version needed when an assistant cites a policy file unrelated to the answer it provides. The lasting guardrail for this specific threat is to validate citation-to-claim evidence and source version, not simply to complete an adjacent service task.
Question 350
During a service audit, a quality assurance lead must accept a change for citation alignment. The issue to be addressed is: An assistant cites a policy file unrelated to the answer it provides. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
To verify this change, the software consultancy should manually inspect citation spans on a difficult evaluation set. Operators still need the implementation to map generated claims to specific retrieved evidence spans and reject unsupported claims and the preventive requirement to validate citation-to-claim evidence and source version.
Incorrect Answers
For multimodal document question answering, it can be sensible to ask a chart trend question and compare the answer to the actual plot if an assistant must reason about a plotted chart and its textual caption. Without the actual change to manually inspect citation spans on a difficult evaluation set, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.
The action to map generated claims to specific retrieved evidence spans and reject unsupported claims can contribute to citation alignment as a implementation step; here, an assistant cites a policy file unrelated to the answer it provides. In the software consultancy acceptance review, trying to map generated claims to specific retrieved evidence spans and reject unsupported claims does not test the exact citation alignment condition where an assistant cites a policy file unrelated to the answer it provides; the relevant test is to manually inspect citation spans on a difficult evaluation set.
In the software consultancy acceptance review, trying to attach a random retrieved URL and call every answer source-grounded does not test the exact citation alignment condition where an assistant cites a policy file unrelated to the answer it provides; the relevant test is to manually inspect citation spans on a difficult evaluation set. Acceptance should instead demonstrate that the service can manually inspect citation spans on a difficult evaluation set, using the described failure as the test criterion.
Developing AI Apps and Agents on Azure (AI-103) Practice Tests
- Microsoft AI-103 Practice Test 33 - 10 Questions (Last Reviewed 20-Oct-2026)
- Microsoft AI-103 Practice Test 34 - 10 Questions (Last Reviewed 19-Oct-2026)
- Microsoft AI-103 Practice Test 35 - 10 Questions (Last Reviewed 18-Oct-2026)
- Microsoft AI-103 Practice Test 36 - 10 Questions (Last Reviewed 17-Oct-2026)
- Microsoft AI-103 Practice Test 37 - 10 Questions (Last Reviewed 16-Oct-2026)