cert
cert-1
cert-2

Pass Microsoft AI-103 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
AI-103 Exam - Verified By Experts
AI-103 Premium File

AI-103 Premium File

$89.99
$98.99
  • Premium File 187 Questions & Answers. Last Update: Nov 15, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$98.99
$89.99
accept 397 downloads in the last 7 days
block-screenshots
Microsoft AI-103 Practice Test 37

Question 360

An invoice has line items in a table and totals beneath it. Which extraction representation helps accounting automation?

  1. Flatten all invoice cells into one string and drop the line-item row-to-amount associations.
  2. Use layout-aware extraction with table rows, columns and numeric fields
  3. Recover reading order based on detected document layout
  4. Use document analysis capable of representing detected layout elements and regions

Explanation

Correct Answer

B:

Flat OCR text cannot reliably preserve which prices belong to which line items. This resolves the specific need to extract invoice table layout and numeric values, while a measure that helps locate extracted document fields in the source addresses a different requirement.

Incorrect Answers

A:

Unstructured text can mix quantities, unit prices and totals. Layout-aware table extraction preserves column and row relationships needed for accurate invoice processing.

C:

Layout-aware reading order is necessary for coherent extracted sentences across columns. This helps when a team needs to preserve multi-column reading sequence; it does not establish how to extract invoice table layout and numeric values.

D:

Non-paragraph marks can be business evidence and need structured representation where supported. The function serves to capture non-textual form elements, whereas the actual requirement is to extract invoice table layout and numeric values.

Question 361

A foundry app processes a one-million-token source but the deployment permits far less context. What should the orchestration layer do?

  1. Place the entire million-token document into one oversized model request.
  2. Use hierarchical retrieval and summarization bounded by model context limits.
  3. Increase the response output-token cap without reducing the source input.
  4. Drop source attribution and randomly sample a few unrelated pages.

Explanation

Correct Answer

B:

No prompt can bypass the serving model's context window; the controller should select and condense relevant evidence.

Incorrect Answers

A:

The deployment's context limit remains binding; an oversized request cannot be processed reliably.

C:

More output tokens do not reduce excessive source context and may worsen the budget.

D:

Ungrounded random truncation can omit relevant evidence rather than preserving the right context.

Question 362

A developer gives an agent Owner on the subscription to allow it to call one model. What security correction is needed?

  1. Assign the agent subscription Owner permanently because its only intended use is model inference.
  2. Reduce the principal's permissions to model-inference roles at the smallest applicable scope
  3. Use authenticated developer identity with a supported credential chain and Azure role assignments
  4. Assign a narrowly scoped Storage Blob Data Reader role to its workload identity

Explanation

Correct Answer

B:

Broad Owner grants are unrelated to the specific model call and unnecessarily enable resource changes. This resolves the specific need to enforce least privilege for model inference, while a measure that helps authenticate automated deployments without client secrets addresses a different requirement.

Incorrect Answers

A:

An intention to call just one model does not constrain the effective subscription Owner privileges. Use a narrowly scoped model-inference authorization instead of broad resource administration.

C:

Developer CLI-based identities can support local token authentication without distributing service keys. This helps when a team needs to use keyless local development credentials; it does not establish how to enforce least privilege for model inference.

D:

Model invocation permission is distinct from reading blobs through the storage data plane. The function serves to permit read-only blob data retrieval, whereas the actual requirement is to enforce least privilege for model inference.

Question 363

An inventory app analyzes 24 fps footage but can process only sparse frames. What must its design account for?

  1. Keep only the first frame from the 24-fps recording and assume it contains every brief inventory event.
  2. Attach segment time ranges to observed claims and extracted visual evidence
  3. Associate extracted frame features with source video IDs and segment offsets
  4. Choose temporal sampling based on event duration and risk of missing brief actions

Explanation

Correct Answer

D:

Sparse sampling can miss transient events, so interval selection must reflect task demands. The decisive difference is whether the design can balance frame processing cost against missed events; the competing function serves to compare observed visual changes through time.

Incorrect Answers

A:

A single frame may miss a short-lived action. Sampling cadence must account for event duration and the false-negative risk of unsampled intervals.

B:

Timestamped evidence allows reviewers to verify what actually happened in video. Here the needed outcome is to balance frame processing cost against missed events; an option designed to ground video summaries in timestamped segments would not directly provide it.

C:

Provenance links allow retrieved frame evidence to be located in the original recording. Its primary purpose is to map search matches back to source video; the stated constraint instead demands a way to balance frame processing cost against missed events.

Question 364

An audio model claims a distressed caller is angry solely based on a short noisy clip. How should confidence be handled?

  1. Infer a caller's mental state as a certainty from noisy speech features alone
  2. Use an audio-capable multimodal reasoning workflow with explicit task criteria
  3. Provide time-coded transcript excerpts and uncertainty flags
  4. Report uncertainty and avoid unsupported personal-state inferences from weak audio

Explanation

Correct Answer

D:

Noisy limited audio does not reliably establish internal emotional states. This directly supports the need to avoid overconfident emotion attribution from audio, unlike a mechanism primarily intended to locate statements by speaker and time.

Incorrect Answers

A:

Noisy audio provides limited evidence about personal emotional state. A model must convey uncertainty and avoid unsupported categorical conclusions.

B:

Voice intent may depend on acoustic as well as lexical evidence; the model should be tested accordingly. That capability is appropriate for interpret spoken intent using audio context, not this separate task of finding how to avoid overconfident emotion attribution from audio.

C:

Grounded transcription references let reviewers revisit contested acoustic evidence. Here the needed outcome is to avoid overconfident emotion attribution from audio; an option designed to trace audio reasoning back to evidence would not directly provide it.

Question 365

A parser returns the correct words from a two-column contract but reads them in the wrong order. What must be configured?

  1. Recover reading order based on detected document layout
  2. Read the two-column contract strictly in raster scan order regardless of column boundaries
  3. Use document analysis capable of representing detected layout elements and regions
  4. Store page numbers and region coordinates for extracted fields

Explanation

Correct Answer

A:

Layout-aware reading order is necessary for coherent extracted sentences across columns. The practical requirement is to preserve multi-column reading sequence; a solution built to extract invoice table layout and numeric values would not meet that same need.

Incorrect Answers

B:

Raster order can splice lines from separate columns. Layout-based reading order is needed to reconstruct each paragraph correctly.

C:

Non-paragraph marks can be business evidence and need structured representation where supported. The function serves to capture non-textual form elements, whereas the actual requirement is to preserve multi-column reading sequence.

D:

Position metadata supports later verification against the original document. That capability is appropriate for locate extracted document fields in the source, not this separate task of finding how to preserve multi-column reading sequence.

Question 366

An assistant can recommend a $50,000 transfer but not execute it without approval. What design is necessary?

  1. Assign the agent unrestricted tenant-wide transfer rights because the user is authenticated to a single account.
  2. Enforce an iteration/time budget with no-progress detection and explicit failure
  3. Require explicit confirmed user intent before privileged destructive actions
  4. Require a verified human approval artifact at the transfer tool boundary

Explanation

Correct Answer

D:

Approval must be enforced by the application handling the real transfer, not only by prompt text. In this scenario the team must gate high-value transfers on human authorization. A feature used to constrain tools to a user's authorized tenant does not substitute for it.

Incorrect Answers

A:

Authentication to an account is not approval to move large sums. A trusted authorization and explicit high-value approval gate must run before any transfer tool executes.

B:

Bounded tool execution prevents runaway use even when the agent cannot self-recognize a loop. Its primary purpose is to terminate runaway autonomous actions; the stated constraint instead demands a way to gate high-value transfers on human authorization.

C:

Irreversible effects should not rely on assumptions about an underspecified user request. That is relevant to confirm intent before destructive operations, but the present scenario requires an ability to gate high-value transfers on human authorization.

Question 367

A local developer wants to avoid using a shared API key while testing Foundry services. What credential mechanism fits?

  1. Check a reusable shared API key into the sample application's repository for local developer convenience.
  2. Use authenticated developer identity with a supported credential chain and Azure role assignments
  3. Verify separate Azure AI Search data-plane authorization for that identity
  4. Assign a narrowly scoped Storage Blob Data Reader role to its workload identity

Explanation

Correct Answer

B:

Developer CLI-based identities can support local token authentication without distributing service keys. The practical requirement is to use keyless local development credentials; a solution built to revoke retired workload access completely would not meet that same need.

Incorrect Answers

A:

A shared secret in source control increases exposure and accountability problems. Use an authenticated developer identity and supported credential chain with appropriately scoped Azure roles.

C:

Project access does not automatically imply permissions to query another service's data. The function serves to authorize search data access separately from project access, whereas the actual requirement is to use keyless local development credentials.

D:

Model invocation permission is distinct from reading blobs through the storage data plane. That capability is appropriate for permit read-only blob data retrieval, not this separate task of finding how to use keyless local development credentials.

Question 368

An agent requests a destructive database operation when user intent is ambiguous. What safeguard is appropriate?

  1. Allow bounded read retries but require idempotency for writes
  2. Change the caller's tenant filter to wildcard and execute a destructive database operation while intent remains uncertain.
  3. Stop side effects while retaining governed logs and trace evidence
  4. Require explicit confirmed user intent before privileged destructive actions

Explanation

Correct Answer

D:

Irreversible effects should not rely on assumptions about an underspecified user request. This resolves the specific need to confirm intent before destructive operations, while a measure that helps differentiate retry safety between reads and writes addresses a different requirement.

Incorrect Answers

A:

Read-only and side-effecting operations have different replay risks and need distinct handling. Its primary purpose is to differentiate retry safety between reads and writes; the stated constraint instead demands a way to confirm intent before destructive operations.

B:

Removing scope restrictions magnifies the consequences of ambiguous intent. Retain least-privilege tenant filtering and require explicit confirmation of the destructive target.

C:

Safe termination should preserve information necessary to diagnose and review the violation. This helps when a team needs to halt unsafe actions without destroying evidence; it does not establish how to confirm intent before destructive operations.

Question 369

A background pipeline deploys resources from CI without keeping service principal passwords. What authentication design fits?

  1. Store a reusable service principal client secret in the build repository so the pipeline can authenticate without rotation.
  2. Use workload identity federation for the build pipeline
  3. Assign a narrowly scoped Storage Blob Data Reader role to its workload identity
  4. Use authenticated developer identity with a supported credential chain and Azure role assignments

Explanation

Correct Answer

B:

Federated tokens allow trusted CI identities to acquire Azure access without static client secrets. Here the relevant capability must let the team authenticate automated deployments without client secrets. An alternative targeting revoke retired workload access completely solves another problem.

Incorrect Answers

A:

A repository secret remains a credential that can be leaked and rotated. Workload identity federation allows the approved CI job to exchange trusted identity assertions instead of distributing long-lived passwords.

C:

Model invocation permission is distinct from reading blobs through the storage data plane. That capability is appropriate for permit read-only blob data retrieval, not this separate task of finding how to authenticate automated deployments without client secrets.

D:

Developer CLI-based identities can support local token authentication without distributing service keys. Here the needed outcome is to authenticate automated deployments without client secrets; an option designed to use keyless local development credentials would not directly provide it.

Question 370

A semi-autonomous workflow may create draft emails but must never send them without review. What state machine pattern fits?

  1. Transition from draft to send only after recorded authorized approval
  2. Permit outbound message-send tools whenever a draft was saved successfully for the same tenant.
  3. Enforce an iteration/time budget with no-progress detection and explicit failure
  4. Allow bounded read retries but require idempotency for writes

Explanation

Correct Answer

A:

A code-enforced approval transition prevents premature outbound communication. The practical requirement is to separate drafting from authorized sending; a solution built to constrain tools to a user's authorized tenant would not meet that same need.

Incorrect Answers

B:

Creating a draft and authorizing delivery are separate transitions. Hold the workflow in awaiting-review and invoke the send tool only after the approved action.

C:

Bounded tool execution prevents runaway use even when the agent cannot self-recognize a loop. The function serves to terminate runaway autonomous actions, whereas the actual requirement is to separate drafting from authorized sending.

D:

Read-only and side-effecting operations have different replay risks and need distinct handling. That capability is appropriate for differentiate retry safety between reads and writes, not this separate task of finding how to separate drafting from authorized sending.

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