Microsoft AI-103 (Developing AI Apps and Agents on Azure) Practice Test 34

Question 330

What test should engineers run after they compare before-and-after equipment photos using visual evidence, rather than relying solely on a successful API response?

  1. Measure false approvals and manual-review precision by confidence band
  2. Check action accuracy and timestamp alignment against annotated segments
  3. Score correct changes, missed changes, and invented differences against annotated image pairs
  4. Measure caption factuality and search usefulness on human-labeled sample images
  5. Review whether users can understand the image's essential message without seeing it

Explanation

Correct Answer

C:

The accepted change is to compare before-and-after equipment photos using visual evidence; this evaluation checks whether it can identify changed switch positions between a baseline photograph and a later inspection photograph, since a multimodal request can include both clearly labeled images and ask for observed differences; single-image captions do not establish what changed across time.

Incorrect Answers

A:

This test concerns the ability to route uncertain visual-extraction results to an operator instead of silently accepting them, for which where supported, confidence estimates and source references enable risk-based human review rather than assuming every extracted field is correct; it does not establish whether the change to compare before-and-after equipment photos using visual evidence works for the current release.

B:

This test concerns the ability to answer what action occurred between defined timestamps in a long maintenance recording, for which temporal segmentation narrows the evidence to the relevant interval and reduces confusion between similar events; it does not establish whether the change to compare before-and-after equipment photos using visual evidence works for the current release.

D:

This test concerns the ability to create short searchable descriptions of individual images without long interpretation, for which captioning summarizes salient visual content for discovery; detailed descriptions and object coordinates are different output requirements; it does not establish whether the change to compare before-and-after equipment photos using visual evidence works for the current release.

E:

This test concerns the ability to describe the meaningful function and visible content of an interface image for screen-reader users, for which alt text prioritizes purpose and essential information, not decorative speculation or excessive narration; human review matters for accessibility; it does not establish whether the change to compare before-and-after equipment photos using visual evidence works for the current release.

Question 331

A solution architect is deciding how to route uncertain visual-extraction results to an operator instead of silently accepting them. Which Azure AI pattern addresses that task?

  1. Use a video analyzer with scene segmentation
  2. Identify object regions in an image
  3. Use a Content Understanding image analyzer
  4. Use confidence and source evidence for review

Explanation

Correct Answer

D:

For the stated requirement to route uncertain visual-extraction results to an operator instead of silently accepting them, where supported, confidence estimates and source references enable risk-based human review rather than assuming every extracted field is correct; this is the mechanism that the system must implement.

Incorrect Answers

A:

For a system whose objective is to classify chapters of a multi-topic video into retrievable units, video analyzers can emit scene segments, timestamps and structured or Markdown results that improve retrieval granularity; this does not directly meet the required outcome to route uncertain visual-extraction results to an operator instead of silently accepting them.

B:

For a system whose objective is to locate components spatially instead of giving only an image-wide caption, region-oriented detection provides object locations or structured references, depending on the model/tool; broad captions do not supply coordinates; this does not directly meet the required outcome to route uncertain visual-extraction results to an operator instead of silently accepting them.

C:

For a system whose objective is to extract selected visual attributes into a repeatable structured schema for an image workflow, an analyzer defines content type, desired fields and models, producing structured evidence for downstream processing; this does not directly meet the required outcome to route uncertain visual-extraction results to an operator instead of silently accepting them.

Question 332

A release reviewer is assessing a feature intended to answer ordered-event questions from video. Which test is most meaningful?

  1. Measure cross-modal event matching on annotated transcript-and-frame pairs
  2. Test schema compliance and classification confusion on representative product images
  3. Score sequence-order judgments against annotated event timelines
  4. Check action accuracy and timestamp alignment against annotated segments

Explanation

Correct Answer

C:

The accepted change is to answer ordered-event questions from video; this evaluation checks whether it can determine whether a safety lock was applied before a machine was started, since temporal reasoning requires preserving sequence across multiple frames; a single still image cannot establish event order.

Incorrect Answers

A:

This test concerns the ability to explain a demonstration whose narration and visible operations carry complementary facts, for which multimodal video understanding aligns spoken words with visual events; ignoring either modality can lose essential meaning; it does not establish whether the change to answer ordered-event questions from video works for the current release.

B:

This test concerns the ability to produce a consistent category and identifier from product photos for a downstream system, for which a declared analyzer schema channels model output into specified field names and types; consumers should still validate values; it does not establish whether the change to answer ordered-event questions from video works for the current release.

D:

This test concerns the ability to answer what action occurred between defined timestamps in a long maintenance recording, for which temporal segmentation narrows the evidence to the relevant interval and reduces confusion between similar events; it does not establish whether the change to answer ordered-event questions from video works for the current release.

Question 333

The planned deployment includes a mechanism to use a Content Understanding image analyzer. Which outcome should the QA team measure?

  1. Measure localization precision and intersection-over-union against annotated regions
  2. Measure retrieval success at the segment level with timestamp ground truth
  3. Test schema compliance and classification confusion on representative product images
  4. Validate extracted field presence and accuracy against labeled inspection images

Explanation

Correct Answer

D:

The accepted change is to use a Content Understanding image analyzer; this evaluation checks whether it can extract selected visual attributes into a repeatable structured schema for an image workflow, since an analyzer defines content type, desired fields and models, producing structured evidence for downstream processing.

Incorrect Answers

A:

This test concerns the ability to locate components spatially instead of giving only an image-wide caption, for which region-oriented detection provides object locations or structured references, depending on the model/tool; broad captions do not supply coordinates; it does not establish whether the change to use a Content Understanding image analyzer works for the current release.

B:

This test concerns the ability to classify chapters of a multi-topic video into retrievable units, for which video analyzers can emit scene segments, timestamps and structured or Markdown results that improve retrieval granularity; it does not establish whether the change to use a Content Understanding image analyzer works for the current release.

C:

This test concerns the ability to produce a consistent category and identifier from product photos for a downstream system, for which a declared analyzer schema channels model output into specified field names and types; consumers should still validate values; it does not establish whether the change to use a Content Understanding image analyzer works for the current release.

Question 334

A user-acceptance test fails with the following observation: A photo-tagging pipeline emits arbitrary labels that cannot be mapped into the catalog database. What is the best correction?

  1. Configure a reusable image analyzer with explicitly defined output fields
  2. Submit both ordered images with baseline and follow-up labels and request grounded observed differences
  3. Enable appropriate temporal segmentation and index segment-level descriptions
  4. Define categories and typed fields for the supported Content Understanding analyzer

Explanation

Correct Answer

D:

The fault signal is that a photo-tagging pipeline emits arbitrary labels that cannot be mapped into the catalog database; the targeted fix works because a declared analyzer schema channels model output into specified field names and types; consumers should still validate values, restoring the ability to produce a consistent category and identifier from product photos for a downstream system.

Incorrect Answers

A:

If the failing function were to extract selected visual attributes into a repeatable structured schema for an image workflow, this change could be relevant because an analyzer defines content type, desired fields and models, producing structured evidence for downstream processing; the actual symptom is that a photo-tagging pipeline emits arbitrary labels that cannot be mapped into the catalog database, which calls for a different repair.

B:

If the failing function were to identify changed switch positions between a baseline photograph and a later inspection photograph, this change could be relevant because a multimodal request can include both clearly labeled images and ask for observed differences; single-image captions do not establish what changed across time; the actual symptom is that a photo-tagging pipeline emits arbitrary labels that cannot be mapped into the catalog database, which calls for a different repair.

C:

If the failing function were to classify chapters of a multi-topic video into retrievable units, this change could be relevant because video analyzers can emit scene segments, timestamps and structured or Markdown results that improve retrieval granularity; the actual symptom is that a photo-tagging pipeline emits arbitrary labels that cannot be mapped into the catalog database, which calls for a different repair.

Question 335

An Azure AI workload passes basic health probes but has this issue: The transcript names a switch, but the assistant cannot tell which on-screen switch the instructor touches. Which fix targets it?

  1. Align timecoded speech and visual frames before generating an evidence-backed description
  2. Generate time-indexed scene descriptions and attach them to retrieval records
  3. Configure a reusable image analyzer with explicitly defined output fields
  4. Retrieve and analyze the frames or clip covering the requested timestamp range

Explanation

Correct Answer

A:

The fault signal is that the transcript names a switch, but the assistant cannot tell which on-screen switch the instructor touches; the targeted fix works because multimodal video understanding aligns spoken words with visual events; ignoring either modality can lose essential meaning, restoring the ability to explain a demonstration whose narration and visible operations carry complementary facts.

Incorrect Answers

B:

If the failing function were to make a long training video discoverable by scenes and visible equipment, this change could be relevant because frame or scene metadata indexes visual evidence that ordinary speech transcripts miss; sampling cadence should match the content; the actual symptom is that the transcript names a switch, but the assistant cannot tell which on-screen switch the instructor touches, which calls for a different repair.

C:

If the failing function were to extract selected visual attributes into a repeatable structured schema for an image workflow, this change could be relevant because an analyzer defines content type, desired fields and models, producing structured evidence for downstream processing; the actual symptom is that the transcript names a switch, but the assistant cannot tell which on-screen switch the instructor touches, which calls for a different repair.

D:

If the failing function were to answer what action occurred between defined timestamps in a long maintenance recording, this change could be relevant because temporal segmentation narrows the evidence to the relevant interval and reduces confusion between similar events; the actual symptom is that the transcript names a switch, but the assistant cannot tell which on-screen switch the instructor touches, which calls for a different repair.

Question 336

Before approving the proposed action to combine speech transcript with video frames, which measurable acceptance check should be required?

  1. Check action accuracy and timestamp alignment against annotated segments
  2. Score sequence-order judgments against annotated event timelines
  3. Score correct changes, missed changes, and invented differences against annotated image pairs
  4. Measure cross-modal event matching on annotated transcript-and-frame pairs

Explanation

Correct Answer

D:

The accepted change is to combine speech transcript with video frames; this evaluation checks whether it can explain a demonstration whose narration and visible operations carry complementary facts, since multimodal video understanding aligns spoken words with visual events; ignoring either modality can lose essential meaning.

Incorrect Answers

A:

This test concerns the ability to answer what action occurred between defined timestamps in a long maintenance recording, for which temporal segmentation narrows the evidence to the relevant interval and reduces confusion between similar events; it does not establish whether the change to combine speech transcript with video frames works for the current release.

B:

This test concerns the ability to determine whether a safety lock was applied before a machine was started, for which temporal reasoning requires preserving sequence across multiple frames; a single still image cannot establish event order; it does not establish whether the change to combine speech transcript with video frames works for the current release.

C:

This test concerns the ability to identify changed switch positions between a baseline photograph and a later inspection photograph, for which a multimodal request can include both clearly labeled images and ask for observed differences; single-image captions do not establish what changed across time; it does not establish whether the change to combine speech transcript with video frames works for the current release.

Question 337

The planned deployment includes a mechanism to identify object regions in an image. Which outcome should the QA team measure?

  1. Measure caption factuality and search usefulness on human-labeled sample images
  2. Measure false approvals and manual-review precision by confidence band
  3. Score correct changes, missed changes, and invented differences against annotated image pairs
  4. Measure localization precision and intersection-over-union against annotated regions

Explanation

Correct Answer

D:

The accepted change is to identify object regions in an image; this evaluation checks whether it can locate components spatially instead of giving only an image-wide caption, since region-oriented detection provides object locations or structured references, depending on the model/tool; broad captions do not supply coordinates.

Incorrect Answers

A:

This test concerns the ability to create short searchable descriptions of individual images without long interpretation, for which captioning summarizes salient visual content for discovery; detailed descriptions and object coordinates are different output requirements; it does not establish whether the change to identify object regions in an image works for the current release.

B:

This test concerns the ability to route uncertain visual-extraction results to an operator instead of silently accepting them, for which where supported, confidence estimates and source references enable risk-based human review rather than assuming every extracted field is correct; it does not establish whether the change to identify object regions in an image works for the current release.

C:

This test concerns the ability to identify changed switch positions between a baseline photograph and a later inspection photograph, for which a multimodal request can include both clearly labeled images and ask for observed differences; single-image captions do not establish what changed across time; it does not establish whether the change to identify object regions in an image works for the current release.

Question 338

A user-acceptance test fails with the following observation: The model detects both actions in separate frames but claims the correct order without temporal evidence. What is the best correction?

  1. Align timecoded speech and visual frames before generating an evidence-backed description
  2. Use a time-aware video analysis sequence and explicitly compare event timestamps
  3. Submit both ordered images with baseline and follow-up labels and request grounded observed differences
  4. Retrieve and analyze the frames or clip covering the requested timestamp range

Explanation

Correct Answer

B:

The fault signal is that the model detects both actions in separate frames but claims the correct order without temporal evidence; the targeted fix works because temporal reasoning requires preserving sequence across multiple frames; a single still image cannot establish event order, restoring the ability to determine whether a safety lock was applied before a machine was started.

Incorrect Answers

A:

If the failing function were to explain a demonstration whose narration and visible operations carry complementary facts, this change could be relevant because multimodal video understanding aligns spoken words with visual events; ignoring either modality can lose essential meaning; the actual symptom is that the model detects both actions in separate frames but claims the correct order without temporal evidence, which calls for a different repair.

C:

If the failing function were to identify changed switch positions between a baseline photograph and a later inspection photograph, this change could be relevant because a multimodal request can include both clearly labeled images and ask for observed differences; single-image captions do not establish what changed across time; the actual symptom is that the model detects both actions in separate frames but claims the correct order without temporal evidence, which calls for a different repair.

D:

If the failing function were to answer what action occurred between defined timestamps in a long maintenance recording, this change could be relevant because temporal segmentation narrows the evidence to the relevant interval and reduces confusion between similar events; the actual symptom is that the model detects both actions in separate frames but claims the correct order without temporal evidence, which calls for a different repair.

Question 339

An Azure AI workload passes basic health probes but has this issue: An automated inspector approves items with weakly supported serial-number readings. Which fix targets it?

  1. Record field evidence and confidence signals, then review low-reliability outputs
  2. Prompt for concise functional alt text and validate it against accessibility criteria
  3. Configure a reusable image analyzer with explicitly defined output fields
  4. Define categories and typed fields for the supported Content Understanding analyzer

Explanation

Correct Answer

A:

The fault signal is that an automated inspector approves items with weakly supported serial-number readings; the targeted fix works because where supported, confidence estimates and source references enable risk-based human review rather than assuming every extracted field is correct, restoring the ability to route uncertain visual-extraction results to an operator instead of silently accepting them.

Incorrect Answers

B:

If the failing function were to describe the meaningful function and visible content of an interface image for screen-reader users, this change could be relevant because alt text prioritizes purpose and essential information, not decorative speculation or excessive narration; human review matters for accessibility; the actual symptom is that an automated inspector approves items with weakly supported serial-number readings, which calls for a different repair.

C:

If the failing function were to extract selected visual attributes into a repeatable structured schema for an image workflow, this change could be relevant because an analyzer defines content type, desired fields and models, producing structured evidence for downstream processing; the actual symptom is that an automated inspector approves items with weakly supported serial-number readings, which calls for a different repair.

D:

If the failing function were to produce a consistent category and identifier from product photos for a downstream system, this change could be relevant because a declared analyzer schema channels model output into specified field names and types; consumers should still validate values; the actual symptom is that an automated inspector approves items with weakly supported serial-number readings, which calls for a different repair.

Question 340

The team plans to use a video analyzer with scene segmentation. Which test provides the clearest evidence that this change works?

  1. Review whether users can understand the image's essential message without seeing it
  2. Measure retrieval success at the segment level with timestamp ground truth
  3. Measure false approvals and manual-review precision by confidence band
  4. Validate extracted field presence and accuracy against labeled inspection images

Explanation

Correct Answer

B:

The accepted change is to use a video analyzer with scene segmentation; this evaluation checks whether it can classify chapters of a multi-topic video into retrievable units, since video analyzers can emit scene segments, timestamps and structured or Markdown results that improve retrieval granularity.

Incorrect Answers

A:

This test concerns the ability to describe the meaningful function and visible content of an interface image for screen-reader users, for which alt text prioritizes purpose and essential information, not decorative speculation or excessive narration; human review matters for accessibility; it does not establish whether the change to use a video analyzer with scene segmentation works for the current release.

C:

This test concerns the ability to route uncertain visual-extraction results to an operator instead of silently accepting them, for which where supported, confidence estimates and source references enable risk-based human review rather than assuming every extracted field is correct; it does not establish whether the change to use a video analyzer with scene segmentation works for the current release.

D:

This test concerns the ability to extract selected visual attributes into a repeatable structured schema for an image workflow, for which an analyzer defines content type, desired fields and models, producing structured evidence for downstream processing; it does not establish whether the change to use a video analyzer with scene segmentation works for the current release.

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