Pass Your Microsoft AI-901 Exam Easy!

Microsoft AI-901 Exam Questions & Answers, Accurate & Verified By IT Experts

Instant Download, Free Fast Updates, 99.6% Pass Rate

AI-901 Premium VCE File

Microsoft AI-901 Premium File

166 Questions & Answers

Last Update: Sep 26, 2026

$89.99

AI-901 Bundle gives you unlimited access to "AI-901" files. However, this does not replace the need for a .vce exam simulator. To download VCE exam simulator click here
AI-901 Premium VCE File
Microsoft AI-901 Premium File

166 Questions & Answers

Last Update: Sep 26, 2026

$89.99

Microsoft AI-901 Exam Bundle gives you unlimited access to "AI-901" files. However, this does not replace the need for a .vce exam simulator. To download your .vce exam simulator click here

Microsoft AI-901 Practice Test 43

Question 420

An archival photo application needs a one-sentence description of an overall scene for search. It does not need individual object boxes. What output should it evaluate?

  1. Image captioning with source-image quality checks and confidence/uncertainty handling.
  2. Only a list of OCR-detected printed letters.
  3. A newly generated stock photograph with a plausible scene.
  4. Bounding boxes without any natural-language scene description.

Explanation

Correct Answer

A:

Captioning summarizes the actual scene in a concise sentence. Validate factual coverage and unsupported-detail rates against the archival image before indexing.

Incorrect Answers

B:

OCR extracts visible letters and cannot describe the scene when the archive needs an overall visual summary.

C:

A newly generated stock image is not the historical source and cannot describe what the original photograph depicts.

D:

Bounding boxes help locate objects but may not provide the requested coherent scene-level description, such as interactions and surroundings.

Question 421

An expenses app extracts an invoice photograph and needs the vendor, tax and line-item totals. A Vision tagger returns "paper" and "receipt." What capability fits the business output?

  1. Generate a summary that estimates the invoice amount from the retailer type.
  2. Document understanding or layout-aware field extraction with typed invoice values and source evidence.
  3. Count the number of "paper" tags as the invoice total.
  4. Perform plain OCR and use the first numeric string near the word "tax" as the invoice total, without field or table validation.

Explanation

Correct Answer

B:

Use an invoice/document extraction analyzer that maps vendor, taxes, amounts and line items to typed fields, table structure and source regions. Verify numerical relationships before committing accounting records.

Incorrect Answers

A:

A guessed invoice amount from retailer identity has no link to the photographed tax, line items or actual charged total. It cannot support financial reconciliation.

C:

Tag counts describe the image category and are unrelated to monetary charges. Two “receipt” tags cannot represent a subtotal or tax.

D:

Raw OCR may find numbers, but proximity does not prove whether an amount is tax, subtotal or final payable total. Typed invoice fields and structural evidence are needed.

Question 422

A Vision API request returns HTTP 401, although the image URL is reachable. What should the developer investigate?

  1. The Vision resource endpoint, credential configuration and effective authorization of the calling principal.
  2. Change the image input resolution and retry with the same invalid Vision service credential.
  3. Generate a new image URL that embeds the caller's claimed job title.
  4. Disable the client's input validation and retry the same secretless call.

Explanation

Correct Answer

A:

HTTP 401 concerns authentication to the Vision service. Verify the resource endpoint, credential or token validity, tenant/region and the caller’s effective permissions independently of public image reachability.

Incorrect Answers

B:

Higher-resolution images may improve recognition after authentication succeeds, but a service-level HTTP 401 is a credential or identity failure. Image quality cannot authorize the caller.

C:

Putting a claimed job title into an image URL has no bearing on the service identity and may leak information without granting authorization.

D:

Input validation does not mint a missing secret or role. Disabling checks while repeating the call increases risk without resolving the 401.

Question 423

A web app uploads a proprietary image format not accepted by the selected vision endpoint. Which validation belongs before sending the request?

  1. Rename the file to `.png` without changing its binary encoding.
  2. Ask the model to ignore unsupported media constraints in the system prompt.
  3. Check actual file encoding, format and size, then convert through an approved supported path if necessary.
  4. Increase output-token limit while keeping invalid input bytes.

Explanation

Correct Answer

C:

Inspect actual magic bytes, format, dimensions and payload size; convert with an approved tool to a supported image representation before inference, preserving task-critical detail.

Incorrect Answers

A:

A renamed file still contains the same proprietary binary payload, which may not be parseable as PNG by the endpoint.

B:

Model prompt text cannot override the media decoder’s supported formats or size limits. This is a client preprocessing requirement.

D:

Output-token limits control generation length, not accepted image encodings. Unsupported bytes must be converted or rejected.

Question 424

An internal inspection app takes private equipment photographs and uploads them to an anonymous public image host solely to supply a model URL. What is safer?

  1. Make the private photo container anonymous for a short upload window, then close public access after the model reads it.
  2. Delete the public link from the final report while keeping the image hosted openly.
  3. Embed internal serial numbers in the public URL to make it easier to match responses.
  4. Use authorized private object access or supported secure byte/image input without making the source publicly discoverable.

Explanation

Correct Answer

D:

Use authenticated private storage, scoped short-lived access or supported secure image bytes where the model accepts them. Keep internal photos out of anonymously public hosts.

Incorrect Answers

A:

Even brief public exposure can disclose confidential equipment photographs and may be cached. Use authenticated object access or supported byte input instead of temporarily removing the access boundary.

B:

Removing a public link from the final report does not revoke direct access to an openly hosted image. The disclosure occurred upstream.

C:

Embedding serials in a URL leaks identifying details through host logs, browser history and link sharing, worsening the confidentiality problem.

Question 425

A claims tool reads a damaged-goods photo and extracts a purported refund amount from a label. The app currently issues the refund immediately. What must be enforced first?

  1. Verify the amount against source image or transaction records and require the caller's refund authorization.
  2. Accept an extraction confidence above the application threshold as sufficient refund approval without checking caller permissions.
  3. Issue the refund then ask the customer to verify the label.
  4. Validate only that the extracted amount has a currency symbol and a parseable number, without reconciling its source or the caller’s refund rights.

Explanation

Correct Answer

A:

Confirm that the amount is read correctly from the source or transaction record and that the caller has refund authority. A vision prediction alone cannot authorize money movement.

Incorrect Answers

B:

Confidence can estimate recognition reliability but cannot grant monetary approval or establish that this customer is entitled to the amount. The transaction service must enforce refund policy and caller authorization.

C:

Paying before validation can create difficult-to-reverse financial losses. Verification and authorization must precede the refund action.

D:

Correct numeric formatting is necessary for processing but does not prove the amount was read from the right label or approved for this claimant. Verify source or transaction records and enforce refund authorization before payment.

Question 426

A warehouse vision system misreads tiny label print after compressing its camera frames. What should the developers check?

  1. Auto-complete missing digits based on inventory frequency.
  2. Source focus, scale, crop and resolution at the model's actual input before adjusting recognition thresholds.
  3. Lower the OCR confidence threshold to accept more compressed-label predictions before checking the actual pixels sent to the model.
  4. Assume a high-quality camera guarantees correct OCR after any downscaling.

Explanation

Correct Answer

B:

Inspect focus, optical scale, crop, compression and actual image resolution reaching the recognizer; preserve enough pixels around the label before changing thresholds. Quality at the model input controls legibility.

Incorrect Answers

A:

Completing missing digits from common inventory patterns can register the wrong physical asset. Frequency is not visual evidence.

C:

A permissive threshold can accept additional wrong digit strings when compression removed source detail. First inspect the actual input resolution and recapture or crop the label; confidence adjustment cannot recover absent evidence.

D:

A high-resolution camera can still yield unreadable model input after aggressive compression. Check the full preprocessing path.

Question 427

An OCR service extracts HTML-looking text printed on a sign, including `<script>` characters. The web client inserts the recognized text directly into the DOM as trusted markup. What is required?

  1. Strip only the exact `<script>` substring from OCR text and inject the remaining string as untrusted HTML.
  2. Assume text is safe because it came from a picture rather than a typed form.
  3. Escape or safely render OCR output as untrusted text rather than executable HTML.
  4. Hide the extracted string in CSS while retaining the same unsafe HTML insertion.

Explanation

Correct Answer

C:

Render recognized strings as escaped text or through safe DOM text APIs, not executable HTML insertion. Treat OCR output as untrusted even when the image itself is legitimate.

Incorrect Answers

A:

A literal substring blacklist does not prevent other executable markup or encoded variants. OCR output must be rendered as text or passed through a robust sanitizer appropriate to the context.

B:

Image origin is not a trust guarantee; attackers can print malicious HTML-like strings on physical signs that OCR reads into the application.

D:

Hiding text via CSS does not prevent inserted script markup from executing. The unsafe parsing behavior remains.

Question 428

A photo-analysis platform handles insurance images from many policyholders. One concurrent request receives another policyholder's result. Which client control is missing?

  1. Correlate every source image, caller authorization and model response to a separate case/session ID.
  2. Remove policyholder names from the UI while allowing incorrect source mixing.
  3. Sort responses by arrival time and assume they match submission order.
  4. Reuse a single global conversation containing all policyholder images.

Explanation

Correct Answer

A:

Assign stable source-image, authenticated case and request identifiers and bind each model response to that exact scope. Concurrent completion order must not cause another policyholder’s inspection findings to appear.

Incorrect Answers

B:

Masking a displayed name does not fix wrong-source attribution or exposure of another claimant’s photo analysis.

C:

Asynchronous requests may finish out of submission order, so sorting only by arrival time cannot preserve the correct case pairing.

D:

A global multimodal conversation can mix unrelated private photographs and customer data. Sessions and retrieval should remain isolated per claimant.

Question 429

An object detector returns five overlapping boxes for three visually similar components. A technician asks about the unit next to a red switch. Select TWO appropriate validation steps.

  1. Resolve duplicate and overlapping detections, preserving coordinates and object confidence.
  2. Choose the highest-confidence component box as the user's intended target without checking overlap or its position relative to the red switch.
  3. Use spatial relationships and the relevant labeled region to bind the technician's referent to the correct object.
  4. Use the first box returned as the target regardless of location.
  5. Sum all boxes and report five distinct components without overlap handling.

Explanation

Correct Answers

A:

Overlapping boxes may represent multiple detections of the same component. Use documented overlap suppression or instance reconciliation while keeping coordinates and confidence evidence.

C:

Locate the red switch in the source image, apply the stated nearest-neighbor/spatial relationship and attach the selected component’s box to the user’s reference.

Incorrect Answers

B:

Detection confidence estimates the likelihood of an object in a box, not whether it is the component the technician described. Resolve overlapping boxes and use the red-switch spatial cue to identify the correct referent.

D:

First-returned detection order is an implementation artifact, not evidence that the component is near the red switch.

E:

Counting all boxes as separate items overstates quantity when detections overlap. Reconcile instances before reporting inventory or targeted findings.

Question 430

A computer-vision endpoint times out after the client uploads a safety inspection image. The UI displays "inspection passed" despite receiving no result. What status handling is appropriate?

  1. Generate a plausible inspection result without analyzing the original image.
  2. Keep an uncertain/pending state, use bounded retries and require a completed source-grounded result before passing inspection.
  3. Retry indefinitely with no client timeout or operator fallback.
  4. Mark safe when the uploaded image file exists.

Explanation

Correct Answer

B:

Preserve pending/failed status, use bounded recoverable retries and require a completed source-grounded inspection result or qualified human review before marking the equipment safe.

Incorrect Answers

A:

Fabricating a safe result after a timeout can permit an uninspected machine to run. The system lacks visual evidence of the guard or defect status.

C:

Infinite retries can hang the interface, consume quota and prevent timely operator intervention. A controlled failure path is essential.

D:

Upload success only proves the file arrived, not that analysis completed or the machinery passed its safety criteria.


Microsoft Azure AI Fundamentals (AI-901) Practice Tests

Purchase Individually

AI-901 Premium File

Premium File
AI-901 Premium File
166 Q&A
$98.99$89.99

Site Search:

 

VISA, MasterCard, AmericanExpress, UnionPay

SPECIAL OFFER: GET 10% OFF

ExamCollection Premium

ExamCollection Premium Files

Pass your Exam with ExamCollection's PREMIUM files!

  • ExamCollection Certified Safe Files
  • Guaranteed to have ACTUAL Exam Questions
  • Up-to-Date Exam Study Material - Verified by Experts
  • Instant Downloads
Enter Your Email Address to Receive Your 10% Off Discount Code
A Confirmation Link will be sent to this email address to verify your login
We value your privacy. We will not rent or sell your email address

SPECIAL OFFER: GET 10% OFF

Use Discount Code:

MIN10OFF

A confirmation link was sent to your e-mail.
Please check your mailbox for a message from support@examcollection.com and follow the directions.

Next

Download Free Demo of VCE Exam Simulator

Experience Avanset VCE Exam Simulator for yourself.

Simply submit your e-mail address below to get started with our interactive software demo of your free trial.

Free Demo Limits: In the demo version you will be able to access only first 5 questions from exam.