Microsoft AI-103 (Developing AI Apps and Agents on Azure) Practice Test 41

Question 400

What test should engineers run after they use LLM translation with contextual instructions, rather than relying solely on a successful API response?

  1. Measure unsafe-content recall and benign-request false-block rate
  2. Compare meaning preservation and style adherence using bilingual human review
  3. Evaluate language identification accuracy on short and code-switched messages
  4. Measure terminology consistency and reviewer corrections across languages

Explanation

Correct Answer

B:

The accepted change is to use LLM translation with contextual instructions; this evaluation checks whether it can translate a conversational response while preserving a specified tone and contextual meaning, since a multilingual model can use conversation context and style constraints; factual meaning and fluent tone need separate verification.

Incorrect Answers

A:

This test concerns the ability to review user messages for policy-relevant hate, violence or self-harm content before generation, for which text safety classification identifies supported content categories; moderation decisions depend on configured policy and context; it does not establish whether the change to use LLM translation with contextual instructions works for the current release.

C:

This test concerns the ability to handle incoming tickets whose language varies and is not declared by the sender, for which language detection supplies a routing signal to a suitable translation or analysis path; short or mixed-language texts need caution; it does not establish whether the change to use LLM translation with contextual instructions works for the current release.

D:

This test concerns the ability to preserve an approved glossary of medical or legal terms across translated responses, for which glossary or controlled terminology guidance reduces inconsistent domain-language choices; blind literal translation may be unsafe; it does not establish whether the change to use LLM translation with contextual instructions works for the current release.

Question 401

The support ticket describes this behavior: A dashboard labels a mixed review neutral and loses the complaint about damaged packaging. Which engineering change should come first?

  1. Extract opinion targets and their associated sentiment rather than only overall polarity
  2. Use key phrase extraction and filter noisy or redundant phrases
  3. Apply structured conversation summarization with action-item and attribution requirements
  4. Apply entity extraction and review entity spans and categories in sample reports

Explanation

Correct Answer

A:

The fault signal is that a dashboard labels a mixed review neutral and loses the complaint about damaged packaging; the targeted fix works because opinion mining links sentiment to specific targets or aspects instead of reporting only a document-level score, restoring the ability to identify that a reviewer likes delivery speed but dislikes package quality in one comment.

Incorrect Answers

B:

If the failing function were to surface the main subjects of short articles without requiring a fixed taxonomy, this change could be relevant because key phrase extraction returns salient terms or phrases; it differs from assigning one of several predefined category labels; the actual symptom is that a dashboard labels a mixed review neutral and loses the complaint about damaged packaging, which calls for a different repair.

C:

If the failing function were to create a handoff note that preserves commitments and unresolved action items from a long support call, this change could be relevant because conversation summaries must capture speakers, decisions and next steps rather than merely extract generic keywords; the actual symptom is that a dashboard labels a mixed review neutral and loses the complaint about damaged packaging, which calls for a different repair.

D:

If the failing function were to identify the people, organizations and places mentioned in incident narratives, this change could be relevant because named entity recognition identifies and classifies mentions in text; it does not by itself establish who authored the document or whether claims are true; the actual symptom is that a dashboard labels a mixed review neutral and loses the complaint about damaged packaging, which calls for a different repair.

Question 402

The planned deployment includes a mechanism to summarize a conversation chronologically. Which outcome should the QA team measure?

  1. Evaluate language identification accuracy on short and code-switched messages
  2. Audit summaries for leakage of labeled private data while checking utility
  3. Evaluate correct action owners, due dates, and decision coverage on labeled calls
  4. Measure unsafe-content recall and benign-request false-block rate

Explanation

Correct Answer

C:

The accepted change is to summarize a conversation chronologically; this evaluation checks whether it can create a handoff note that preserves commitments and unresolved action items from a long support call, since conversation summaries must capture speakers, decisions and next steps rather than merely extract generic keywords.

Incorrect Answers

A:

This test concerns the ability to handle incoming tickets whose language varies and is not declared by the sender, for which language detection supplies a routing signal to a suitable translation or analysis path; short or mixed-language texts need caution; it does not establish whether the change to summarize a conversation chronologically works for the current release.

B:

This test concerns the ability to produce a useful incident synopsis without retaining unnecessary personal identifiers, for which a redaction stage can remove sensitive spans from the generated artifact; summarization alone does not guarantee privacy; it does not establish whether the change to summarize a conversation chronologically works for the current release.

D:

This test concerns the ability to review user messages for policy-relevant hate, violence or self-harm content before generation, for which text safety classification identifies supported content categories; moderation decisions depend on configured policy and context; it does not establish whether the change to summarize a conversation chronologically works for the current release.

Question 403

After deployment, the team observes: A routing engine finds customer names but cannot decide whether a request concerns billing or shipping. Which remediation addresses the likely cause?

  1. Apply sentiment analysis with representative domain testing and interpret mixed cases carefully
  2. Use a strict output schema and application validation for all expected fields
  3. Apply PII detection and redact or tokenize the detected fields before logging
  4. Apply supervised or prompted classification with explicit allowed labels

Explanation

Correct Answer

D:

The fault signal is that a routing engine finds customer names but cannot decide whether a request concerns billing or shipping; the targeted fix works because a classifier assigns labels from an agreed taxonomy; named entity extraction alone does not decide message intent, restoring the ability to route written requests into one of several approved service categories.

Incorrect Answers

A:

If the failing function were to detect broadly positive, negative or neutral attitudes expressed in reviews, this change could be relevant because sentiment analysis estimates evaluative polarity in text; it should not be used as a precise diagnosis of the reviewer's emotions; the actual symptom is that a routing engine finds customer names but cannot decide whether a request concerns billing or shipping, which calls for a different repair.

B:

If the failing function were to extract account ID, reason category, and decision text into fixed machine-consumable fields, this change could be relevant because structured output controls field names and types for downstream code; domain validation must still check allowed values; the actual symptom is that a routing engine finds customer names but cannot decide whether a request concerns billing or shipping, which calls for a different repair.

C:

If the failing function were to find private identifiers in user-provided text so they can be minimized or masked, this change could be relevant because PII recognition identifies supported sensitive spans for redaction or controlled processing; keeping the raw text is a separate policy decision; the actual symptom is that a routing engine finds customer names but cannot decide whether a request concerns billing or shipping, which calls for a different repair.

Question 404

A design review proposes to flag potentially harmful text in a moderation flow. Which measurable check would establish success on real inputs?

  1. Audit summaries for leakage of labeled private data while checking utility
  2. Evaluate language identification accuracy on short and code-switched messages
  3. Measure factual consistency and exception recall against annotated document sections
  4. Measure unsafe-content recall and benign-request false-block rate

Explanation

Correct Answer

D:

The accepted change is to flag potentially harmful text in a moderation flow; this evaluation checks whether it can review user messages for policy-relevant hate, violence or self-harm content before generation, since text safety classification identifies supported content categories; moderation decisions depend on configured policy and context.

Incorrect Answers

A:

This test concerns the ability to produce a useful incident synopsis without retaining unnecessary personal identifiers, for which a redaction stage can remove sensitive spans from the generated artifact; summarization alone does not guarantee privacy; it does not establish whether the change to flag potentially harmful text in a moderation flow works for the current release.

B:

This test concerns the ability to handle incoming tickets whose language varies and is not declared by the sender, for which language detection supplies a routing signal to a suitable translation or analysis path; short or mixed-language texts need caution; it does not establish whether the change to flag potentially harmful text in a moderation flow works for the current release.

C:

This test concerns the ability to condense a long policy document while preserving key exceptions and source support, for which grounded summarization uses supplied text as its evidence; a fluent summary must not introduce unsupported policy claims; it does not establish whether the change to flag potentially harmful text in a moderation flow works for the current release.

Question 405

A solution architect is deciding how to route written requests into one of several approved service categories. Which Azure AI pattern addresses that task?

  1. Request schema-constrained entity JSON
  2. Classify messages against a defined taxonomy
  3. Detect sensitive personal information before reuse
  4. Redact private details from summarized text

Explanation

Correct Answer

B:

For the stated requirement to route written requests into one of several approved service categories, a classifier assigns labels from an agreed taxonomy; named entity extraction alone does not decide message intent; this is the mechanism that the system must implement.

Incorrect Answers

A:

For a system whose objective is to extract account ID, reason category, and decision text into fixed machine-consumable fields, structured output controls field names and types for downstream code; domain validation must still check allowed values; this does not directly meet the required outcome to route written requests into one of several approved service categories.

C:

For a system whose objective is to find private identifiers in user-provided text so they can be minimized or masked, PII recognition identifies supported sensitive spans for redaction or controlled processing; keeping the raw text is a separate policy decision; this does not directly meet the required outcome to route written requests into one of several approved service categories.

D:

For a system whose objective is to produce a useful incident synopsis without retaining unnecessary personal identifiers, a redaction stage can remove sensitive spans from the generated artifact; summarization alone does not guarantee privacy; this does not directly meet the required outcome to route written requests into one of several approved service categories.

Question 406

The support ticket describes this behavior: A parser breaks because generated account summaries inconsistently use prose and alternate property names. Which engineering change should come first?

  1. Use a strict output schema and application validation for all expected fields
  2. Apply supervised or prompted classification with explicit allowed labels
  3. Supply precise domain definitions and representative labeled edge cases
  4. Use key phrase extraction and filter noisy or redundant phrases

Explanation

Correct Answer

A:

The fault signal is that a parser breaks because generated account summaries inconsistently use prose and alternate property names; the targeted fix works because structured output controls field names and types for downstream code; domain validation must still check allowed values, restoring the ability to extract account ID, reason category, and decision text into fixed machine-consumable fields.

Incorrect Answers

B:

If the failing function were to route written requests into one of several approved service categories, this change could be relevant because a classifier assigns labels from an agreed taxonomy; named entity extraction alone does not decide message intent; the actual symptom is that a parser breaks because generated account summaries inconsistently use prose and alternate property names, which calls for a different repair.

C:

If the failing function were to extract a specialized compliance category when generic terminology is overloaded, this change could be relevant because task-specific definitions and examples help a model map ambiguous words to the intended domain meaning; the actual symptom is that a parser breaks because generated account summaries inconsistently use prose and alternate property names, which calls for a different repair.

D:

If the failing function were to surface the main subjects of short articles without requiring a fixed taxonomy, this change could be relevant because key phrase extraction returns salient terms or phrases; it differs from assigning one of several predefined category labels; the actual symptom is that a parser breaks because generated account summaries inconsistently use prose and alternate property names, which calls for a different repair.

Question 407

A new production AI solution must review user messages for policy-relevant hate, violence or self-harm content before generation. Which approach should the team select first?

  1. Generate a grounded document summary
  2. Redact private details from summarized text
  3. Use opinion mining for aspect-level sentiment
  4. Detect source language before routing text
  5. Flag potentially harmful text in a moderation flow

Explanation

Correct Answer

E:

For the stated requirement to review user messages for policy-relevant hate, violence or self-harm content before generation, text safety classification identifies supported content categories; moderation decisions depend on configured policy and context; this is the mechanism that the system must implement.

Incorrect Answers

A:

For a system whose objective is to condense a long policy document while preserving key exceptions and source support, grounded summarization uses supplied text as its evidence; a fluent summary must not introduce unsupported policy claims; this does not directly meet the required outcome to review user messages for policy-relevant hate, violence or self-harm content before generation.

B:

For a system whose objective is to produce a useful incident synopsis without retaining unnecessary personal identifiers, a redaction stage can remove sensitive spans from the generated artifact; summarization alone does not guarantee privacy; this does not directly meet the required outcome to review user messages for policy-relevant hate, violence or self-harm content before generation.

C:

For a system whose objective is to identify that a reviewer likes delivery speed but dislikes package quality in one comment, opinion mining links sentiment to specific targets or aspects instead of reporting only a document-level score; this does not directly meet the required outcome to review user messages for policy-relevant hate, violence or self-harm content before generation.

D:

For a system whose objective is to handle incoming tickets whose language varies and is not declared by the sender, language detection supplies a routing signal to a suitable translation or analysis path; short or mixed-language texts need caution; this does not directly meet the required outcome to review user messages for policy-relevant hate, violence or self-harm content before generation.

Question 408

The implementation does not meet its business requirement: An analyst uses general keyword matching and misses alternate surface forms of organization names. Which corrective action is appropriate?

  1. Use key phrase extraction and filter noisy or redundant phrases
  2. Extract opinion targets and their associated sentiment rather than only overall polarity
  3. Translate with context and explicit tone requirements, then assess semantic fidelity
  4. Apply entity extraction and review entity spans and categories in sample reports

Explanation

Correct Answer

D:

The fault signal is that an analyst uses general keyword matching and misses alternate surface forms of organization names; the targeted fix works because named entity recognition identifies and classifies mentions in text; it does not by itself establish who authored the document or whether claims are true, restoring the ability to identify the people, organizations and places mentioned in incident narratives.

Incorrect Answers

A:

If the failing function were to surface the main subjects of short articles without requiring a fixed taxonomy, this change could be relevant because key phrase extraction returns salient terms or phrases; it differs from assigning one of several predefined category labels; the actual symptom is that an analyst uses general keyword matching and misses alternate surface forms of organization names, which calls for a different repair.

B:

If the failing function were to identify that a reviewer likes delivery speed but dislikes package quality in one comment, this change could be relevant because opinion mining links sentiment to specific targets or aspects instead of reporting only a document-level score; the actual symptom is that an analyst uses general keyword matching and misses alternate surface forms of organization names, which calls for a different repair.

C:

If the failing function were to translate a conversational response while preserving a specified tone and contextual meaning, this change could be relevant because a multilingual model can use conversation context and style constraints; factual meaning and fluent tone need separate verification; the actual symptom is that an analyst uses general keyword matching and misses alternate surface forms of organization names, which calls for a different repair.

Question 409

A Foundry-based application is being scoped for production. The non-negotiable task is to create a handoff note that preserves commitments and unresolved action items from a long support call. Which implementation fits?

  1. Generate a grounded document summary
  2. Summarize a conversation chronologically
  3. Flag potentially harmful text in a moderation flow
  4. Use LLM translation with contextual instructions

Explanation

Correct Answer

B:

For the stated requirement to create a handoff note that preserves commitments and unresolved action items from a long support call, conversation summaries must capture speakers, decisions and next steps rather than merely extract generic keywords; this is the mechanism that the system must implement.

Incorrect Answers

A:

For a system whose objective is to condense a long policy document while preserving key exceptions and source support, grounded summarization uses supplied text as its evidence; a fluent summary must not introduce unsupported policy claims; this does not directly meet the required outcome to create a handoff note that preserves commitments and unresolved action items from a long support call.

C:

For a system whose objective is to review user messages for policy-relevant hate, violence or self-harm content before generation, text safety classification identifies supported content categories; moderation decisions depend on configured policy and context; this does not directly meet the required outcome to create a handoff note that preserves commitments and unresolved action items from a long support call.

D:

For a system whose objective is to translate a conversational response while preserving a specified tone and contextual meaning, a multilingual model can use conversation context and style constraints; factual meaning and fluent tone need separate verification; this does not directly meet the required outcome to create a handoff note that preserves commitments and unresolved action items from a long support call.

Question 410

A cloud engineering team must detect broadly positive, negative or neutral attitudes expressed in reviews. What is the most appropriate Azure AI implementation?

  1. Analyze sentiment in customer feedback
  2. Use opinion mining for aspect-level sentiment
  3. Extract named entities from reports
  4. Detect source language before routing text

Explanation

Correct Answer

A:

For the stated requirement to detect broadly positive, negative or neutral attitudes expressed in reviews, sentiment analysis estimates evaluative polarity in text; it should not be used as a precise diagnosis of the reviewer's emotions; this is the mechanism that the system must implement.

Incorrect Answers

B:

For a system whose objective is to identify that a reviewer likes delivery speed but dislikes package quality in one comment, opinion mining links sentiment to specific targets or aspects instead of reporting only a document-level score; this does not directly meet the required outcome to detect broadly positive, negative or neutral attitudes expressed in reviews.

C:

For a system whose objective is to identify the people, organizations and places mentioned in incident narratives, named entity recognition identifies and classifies mentions in text; it does not by itself establish who authored the document or whether claims are true; this does not directly meet the required outcome to detect broadly positive, negative or neutral attitudes expressed in reviews.

D:

For a system whose objective is to handle incoming tickets whose language varies and is not declared by the sender, language detection supplies a routing signal to a suitable translation or analysis path; short or mixed-language texts need caution; this does not directly meet the required outcome to detect broadly positive, negative or neutral attitudes expressed in reviews.

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