McAfee Secure

Exam Code: AI-901

Exam Name: Microsoft Azure AI Fundamentals

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Fundamentals

Microsoft AI-901 Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

166 Questions & Answers with Testing Engine

"Microsoft Azure AI Fundamentals Exam", also known as AI-901 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-901 Exam Engine. Your AI-901 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-901 Practice Test 46

Question 450

An Azure AI Fundamentals solution has this requirement: An agent must answer only from company-approved product manuals. Which implementation most directly addresses it?

  1. Choose a model size using representative quality and cost benchmarks
  2. Set authorized knowledge sources and a no-source fallback
  3. Handle tool timeout as a typed error and preserve partial workflow state
  4. Externalize endpoint and deployment configuration per environment

Explanation

Correct Answer

B:

Retrieval access and abstention avoid inventing unsupported documentation. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Lower-cost deployment is acceptable only if accuracy and safety floors remain met. That is not sufficient here: Retrieval access and abstention avoid inventing unsupported documentation.

C:

Timeout handling should not be reported as a successful action. That is not sufficient here: Retrieval access and abstention avoid inventing unsupported documentation.

D:

A correct code path can still target the wrong resource after migration. That is not sufficient here: Retrieval access and abstention avoid inventing unsupported documentation.

Question 451

Troubleshooting report: The app guesses the tax amount when the photo cuts it off. Which correction is MOST directly targeted to this failure?

  1. Analyze multiple temporal segments and preserve timecodes
  2. Treat absent fields as unknown and request a better image
  3. Detect people and helmets and evaluate spatial associations
  4. Require a user-approved booking and payment action

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Images of receipts need OCR/layout extraction before typed business values. The targeted change is: Treat absent fields as unknown and request a better image.

Incorrect Answers

A:

Video understanding retains order and time, unlike single-image tagging. In the stated case, troubleshooting report: The app guesses the tax amount when the photo cuts it off; “Analyze multiple temporal segments and preserve timecodes” does not target that failure.

C:

Detection relates objects to positions for compliance checks. In the stated case, troubleshooting report: The app guesses the tax amount when the photo cuts it off; “Detect people and helmets and evaluate spatial associations” does not target that failure.

D:

An agent may propose actions but payment must remain permission-controlled. In the stated case, troubleshooting report: The app guesses the tax amount when the photo cuts it off; “Require a user-approved booking and payment action” does not target that failure.

Question 452

Troubleshooting report: The system retains only the first line item. Which correction is MOST directly targeted to this failure?

  1. Represent repeated fields as arrays and validate counts
  2. Build citations from stored IDs and source spans
  3. Run deterministic line-item reconciliation before acceptance
  4. Preserve segment timestamps and review rapid events separately when sampling might miss them

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. Repeated records require structured lists with item-level associations. The targeted change is: Represent repeated fields as arrays and validate counts.

Incorrect Answers

B:

Verifiable provenance differentiates a cited answer from guesswork. In the stated case, troubleshooting report: The system retains only the first line item; “Build citations from stored IDs and source spans” does not target that failure.

C:

Independent business rules catch plausible extraction errors. In the stated case, troubleshooting report: The system retains only the first line item; “Run deterministic line-item reconciliation before acceptance” does not target that failure.

D:

Temporal evidence supports the timing of observed events, but approximately one frame per second cannot guarantee subsecond event capture. In the stated case, troubleshooting report: The system retains only the first line item; “Preserve segment timestamps and review rapid events separately when sampling might miss them” does not target that failure.

Question 453

An acceptance tester needs to validate: Keep recognized visual labels separate from text-read lines and words. Which observed result proves the intended behavior?

  1. A multi-object test returns boxes when location is required
  2. Holdout images correctly separate the two defect types
  3. A returned serial number is stored as OCR text not a generic image tag
  4. A test image receives relevant scene and object tags

Explanation

Correct Answer

C:

Tags describe concepts while OCR reports character content. The expected acceptance evidence is: A returned serial number is stored as OCR text not a generic image tag. This observation directly tests the stated behavior.

Incorrect Answers

A:

Classification predicts image-level categories while detection provides positions. This observation does not establish the required result: A returned serial number is stored as OCR text not a generic image tag.

B:

Fine-grained visual classes require representative discriminating images. This observation does not establish the required result: A returned serial number is stored as OCR text not a generic image tag.

D:

Tags summarize recognized concepts as labels. This observation does not establish the required result: A returned serial number is stored as OCR text not a generic image tag.

Question 454

An engineering review documents this requirement: An app wants to crop a picture automatically to emphasize the main subject. Select the most appropriate solution.

  1. Use a supported smart-crop feature with the requested aspect ratio
  2. Request an image caption or supported multimodal description
  3. Use a supported dense-captioning capability where available
  4. Provide the actual image to a capable multimodal or OCR pipeline

Explanation

Correct Answer

A:

Smart cropping attempts to retain salient regions within target dimensions. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Captioning produces a sentence-level description rather than a coordinate list. That is not sufficient here: Smart cropping attempts to retain salient regions within target dimensions.

C:

Dense captions describe localized regions beyond a single global caption. That is not sufficient here: Smart cropping attempts to retain salient regions within target dimensions.

D:

Visual evidence is needed to read sign content. That is not sufficient here: Smart cropping attempts to retain salient regions within target dimensions.

Question 455

A project needs the following AI behavior: A vision service must label the main category of an image. What should the designer implement?

  1. Use an image-classification model with class probabilities
  2. Choose a model or pipeline that supports both visual and audio modalities
  3. Explain token-based next-token prediction conditioned on preceding context
  4. Review precision, recall and the confusion matrix rather than accuracy alone

Explanation

Correct Answer

A:

Classification assigns image-wide categories rather than bounding-box coordinates. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Different media channels provide complementary scene and dialogue evidence. That is not sufficient here: Classification assigns image-wide categories rather than bounding-box coordinates.

C:

A generative language model estimates likely continuations rather than reading a stored answer paragraph. That is not sufficient here: Classification assigns image-wide categories rather than bounding-box coordinates.

D:

High accuracy can conceal failure to detect rare positive events. That is not sufficient here: Classification assigns image-wide categories rather than bounding-box coordinates.

Question 456

An Azure-based AI team must satisfy this objective: An AI app needs to run in a different tenant with separate credentials. Which design choice is best?

  1. Use tenant-specific identity, endpoint configuration and permissions
  2. Write an instruction to identify missing inputs and request clarification before tools
  3. Compare model deployment, prompt version, retrieval index and tool configuration
  4. Send the image in a supported multimodal message along with the question

Explanation

Correct Answer

A:

Copying code does not grant access to another tenant's resources. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Prompts can define required preconditions but tools must still enforce them. That is not sufficient here: Copying code does not grant access to another tenant's resources.

C:

Environment drift can change behavior without code edits. That is not sufficient here: Copying code does not grant access to another tenant's resources.

D:

The model must receive the actual visual content to reason about it. That is not sufficient here: Copying code does not grant access to another tenant's resources.

Question 457

Troubleshooting report: The solution writes an unconstrained paragraph instead of a category. Which correction is MOST directly targeted to this failure?

  1. Use classification outputs for the entire image
  2. Train or configure intent classification with a bounded label set
  3. Choose segmentation with validated pixel-level damage regions
  4. Use entity recognition and verify sensitive-entity recall

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Classification maps text to one of several predefined categories. The targeted change is: Train or configure intent classification with a bounded label set.

Incorrect Answers

A:

Image classification predicts image-level labels. In the stated case, troubleshooting report: The solution writes an unconstrained paragraph instead of a category; “Use classification outputs for the entire image” does not target that failure.

C:

Pixel-level region masks localize building damage more precisely than an overall image class label. In the stated case, troubleshooting report: The solution writes an unconstrained paragraph instead of a category; “Choose segmentation with validated pixel-level damage regions” does not target that failure.

D:

Entity extraction locates sensitive spans for privacy handling. In the stated case, troubleshooting report: The solution writes an unconstrained paragraph instead of a category; “Use entity recognition and verify sensitive-entity recall” does not target that failure.

Question 458

An Azure-based AI team must satisfy this objective: An assistant must answer from confidential contract excerpts after extraction. Which design choice is best?

  1. Apply authorized retrieval filters to extracted chunks and cited source text
  2. Classify document type before routing to the relevant extraction analyzer
  3. Check feature, model and regional availability before creating an analyzer
  4. Preserve page identifiers and validate page sequence during extraction

Explanation

Correct Answer

A:

Extracting content does not waive document access controls. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Different document kinds have distinct field schemas and review paths. That is not sufficient here: Extracting content does not waive document access controls.

C:

Some modalities and features depend on resource and API version. That is not sufficient here: Extracting content does not waive document access controls.

D:

Page order and references are part of document provenance. That is not sufficient here: Extracting content does not waive document access controls.

Question 459

Troubleshooting report: The versions are tested on different recordings and compared naively. Which correction is MOST directly targeted to this failure?

  1. Add an authorized synthesis stage for the answer
  2. Use a fixed annotated evaluation set and consistent metrics
  3. Add a no-speech branch and clarification option
  4. Select the correct source/target language translation path

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Comparable labeled audio prevents misleading quality claims. The targeted change is: Use a fixed annotated evaluation set and consistent metrics.

Incorrect Answers

A:

Input transcription and output synthesis are distinct steps. In the stated case, troubleshooting report: The versions are tested on different recordings and compared naively; “Add an authorized synthesis stage for the answer” does not target that failure.

C:

An empty recognition result is not semantically equivalent to the word no. In the stated case, troubleshooting report: The versions are tested on different recordings and compared naively; “Add a no-speech branch and clarification option” does not target that failure.

D:

Translation requires semantic equivalence and language direction. In the stated case, troubleshooting report: The versions are tested on different recordings and compared naively; “Select the correct source/target language translation path” does not target that failure.

Question 460

An acceptance tester needs to validate: Use operation IDs and read-back receipts before replaying external side effects. Which observed result proves the intended behavior?

  1. A trace shows which tool result supported each claimed business action
  2. Execution traces show the intended tool invoked with valid arguments
  3. Repeated events yield at most one order creation under the same operation ID
  4. A repository scan reveals no hardcoded secret and the app still authenticates

Explanation

Correct Answer

C:

A timeout does not prove that the original mutation failed. The expected acceptance evidence is: Repeated events yield at most one order creation under the same operation ID. This observation directly tests the stated behavior.

Incorrect Answers

A:

Operational observability aids debugging without exposing hidden sensitive prompt data. This observation does not establish the required result: Repeated events yield at most one order creation under the same operation ID.

B:

Tools must be discoverable with valid schemas before selection. This observation does not establish the required result: Repeated events yield at most one order creation under the same operation ID.

D:

Credentials should not appear in repositories, prompt logs or client binaries. This observation does not establish the required result: Repeated events yield at most one order creation under the same operation ID.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests