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 25

Question 240

An enrichment pipeline needs a proprietary classifier not supplied by built-in skills. What extensibility mechanism fits?

  1. Represent a custom business classification as an unsupported built-in enrichment function
  2. Configure entity extraction as an enrichment step before indexing
  3. Invoke a custom enrichment skill with a defined request and response contract
  4. Add document-field or key-value extraction using supported content analyzers

Explanation

Correct Answer

C:

Custom skills insert application-owned processing into the ingestion workflow. The decisive difference is whether the design can extend indexing with custom classification; the competing function serves to map enrichment outputs into indexed fields.

Incorrect Answers

A:

An indexer cannot execute arbitrary custom logic by naming a nonexistent built-in skill. A supported custom skill endpoint and proper output mapping are needed.

B:

Index-time enrichment adds structured searchable fields derived from source content. Here the needed outcome is to extend indexing with custom classification; an option designed to enrich index records with extracted entities would not directly provide it.

D:

Field-aware extraction makes important structured document attributes accessible to filters or agents. That is relevant to index key-value form fields, but the present scenario requires an ability to extend indexing with custom classification.

Question 241

An AI receptionist must remember a user's preferred language across sessions with consent. Where should this be stored?

  1. Store the user's preferred language as an unrestricted tool-permission grant shared with all users.
  2. Store approved durable preferences separately from ephemeral working conversation state
  3. Maintain a short ordered conversation and approval state tied to the task
  4. Expire stale task context and retrieve current policy before sensitive decisions

Explanation

Correct Answer

B:

Persistent preferences have different lifecycle and privacy requirements than transient chat memory. Here the relevant capability must let the team retain permitted long-term user preference data. An alternative targeting constrain an agent to an authorized task solves another problem.

Incorrect Answers

A:

An approved language preference is profile data, not authorization. Store it in a consented user-scoped memory record separate from tool privileges.

C:

Relevant conversation state keeps authorization and progress coherent across tool cycles. That capability is appropriate for retain task-scoped user decisions during turns, not this separate task of finding how to retain permitted long-term user preference data.

D:

Short-lived task facts should not override newer authoritative policy evidence. Here the needed outcome is to retain permitted long-term user preference data; an option designed to limit stale memory from overruling current rules would not directly provide it.

Question 242

A regulated pipeline must avoid hardcoded credentials, verify behavior before promotion, and require review of safety-rule changes. Which THREE controls serve those needs? Choose THREE.

  1. Store immutable versioned prompt and model deployment manifests for rollback
  2. Use workload identity or managed federation rather than embedded pipeline secrets
  3. Apply infrastructure changes directly to production with no version control, tests, or recorded approval path.
  4. Require a recorded protected-environment approval gate for safety changes
  5. Assume production tool permissions match staging and omit runtime permission tests.
  6. Run a representative evaluation suite with defined failure thresholds before promotion

Explanation

Correct Answers

B:

Federated or managed identities avoid long-lived credentials committed to repositories. Quality tests, secretless deployment identity and a recorded environment approval gate are independent safeguards.

D:

Release approvals must be checked by the pipeline, not just requested in a prompt.

F:

A measurable evaluation gate prevents known regressions from reaching users.

Incorrect Answers

A:

Recovering the last known-good prompt/model combination requires retaining those exact artifacts. Its suitable role is to restore a known working prompt-model pair; it would leave the specific need to validate quality, use secretless CI auth, and enforce safety approvals unresolved.

C:

Untracked infrastructure changes do not verify behavior or enforce the protected safety-rule approval required by the regulated pipeline.

E:

An unverified production permission assumption cannot demonstrate that the deployment will work safely after promotion.

Question 243

A carousel contains four photographs showing different steps in a repair. What should the captions communicate?

  1. Write the identical caption “repair step” under all four photos without identifying what changes at each step.
  2. Write context-specific descriptions of each step rather than repeating one generic caption
  3. Review generated alternatives against the final published image versions
  4. Reuse one generic caption for every photograph despite the different repair steps shown.

Explanation

Correct Answer

B:

Per-image alternatives let nonvisual users follow the sequence. The practical requirement is to distinguish meaningful steps across several images; a solution built to avoid speculative attributes in alt text would not meet that same need.

Incorrect Answers

A:

Generic repeated captions remove the step-by-step distinctions users need. Describe the meaningful action visible in each photograph.

C:

Text alternatives must correspond to what users actually see after edits or revisions. The function serves to keep accessibility descriptions aligned with final media, whereas the actual requirement is to distinguish meaningful steps across several images.

D:

Identical generic captions lose the step-by-step information. Each caption should identify the distinct action or result visible in its photograph.

Question 244

A model returns valid JSON for claims but invents missing deductible values. What rule must be added?

  1. Represent unsupported fields as null or unavailable rather than fabricating values
  2. Populate an absent deductible with the model's most probable amount
  3. Use representative domain reports with explicitly labeled exception examples
  4. Add modality-sensitive domain examples distinguishing permissions from obligations

Explanation

Correct Answer

A:

Extractive business applications must not invent values absent from the source. The practical requirement is to represent missing evidence explicitly in domain extraction; a solution built to extract actionable compliance obligations would not meet that same need.

Incorrect Answers

B:

A well-formed JSON object can still contain fabricated financial values. Missing evidence should remain unknown or trigger review rather than be guessed.

C:

Domain evidence and exception cases let teams test whether critical caveats survive summarization. The function serves to evaluate finance-specific exception handling, whereas the actual requirement is to represent missing evidence explicitly in domain extraction.

D:

The difference between optional and mandatory terms is material to legal task accuracy. That capability is appropriate for interpret domain-specific modal language, not this separate task of finding how to represent missing evidence explicitly in domain extraction.

Question 245

A skillset produces an output field but it never appears in the target search index. What should be checked?

  1. Assume a skillset output becomes searchable without creating a target field mapping
  2. Invoke a custom enrichment skill with a defined request and response contract
  3. Verify output field mappings and target index schema compatibility
  4. Add document-field or key-value extraction using supported content analyzers

Explanation

Correct Answer

C:

Enrichment results must be explicitly mapped to correctly typed index fields. In this scenario the team must map enrichment outputs into indexed fields. A feature used to enrich index records with extracted entities does not substitute for it.

Incorrect Answers

A:

An enrichment result is not necessarily written to an index field automatically. The configured output and field mappings must match the target schema.

B:

Custom skills insert application-owned processing into the ingestion workflow. Its primary purpose is to extend indexing with custom classification; the stated constraint instead demands a way to map enrichment outputs into indexed fields.

D:

Field-aware extraction makes important structured document attributes accessible to filters or agents. This helps when a team needs to index key-value form fields; it does not establish how to map enrichment outputs into indexed fields.

Question 246

An agent summarizes a long conversation and mistakenly removes a customer's pending approval requirement. What must the memory summary preserve?

  1. Retain only a cheerful sentiment summary of the previous conversation.
  2. Retain active constraints, unfulfilled approvals, task state and provenance during context compaction.
  3. Delete every unresolved approval after a fixed number of turns.
  4. Copy the full transcript into every prompt without selecting active constraints.

Explanation

Correct Answer

B:

Compression should not discard authorization conditions needed for subsequent actions.

Incorrect Answers

A:

Sentiment does not preserve the pending approval or current workflow requirements.

C:

Age-based pruning can remove active authorizations and cause an unintended side effect.

D:

Unbounded transcript replay grows context cost and can reintroduce stale instructions.

Question 247

An API frequently returns throttling responses when many customers submit prompts at once, but average response time is acceptable at lower load. What should an operator inspect first?

  1. Inspect model deployment throughput limits and subscription quota consumption
  2. Shorten completed model responses but ignore 429 response headers and request concurrency limits.
  3. Validate regional model quota and request capacity changes before cutover
  4. Estimate input tokens, output tokens, retrieval calls and forecast traffic volume

Explanation

Correct Answer

A:

Request-per-minute and token-per-minute limits can throttle bursts independently of underlying application server capacity. The decisive difference is whether the design can diagnose serving quota exhaustion under bursts; the competing function serves to bound output token usage.

Incorrect Answers

B:

Smaller outputs may reduce token usage, but burst throttling requires measuring request/token rate and handling provider quota, retry-after signals and client concurrency at the deployment boundary.

C:

Quotas are resource and deployment constraints; capacity should be confirmed before directing production traffic. Its primary purpose is to prepare regional serving capacity; the stated constraint instead demands a way to diagnose serving quota exhaustion under bursts.

D:

AI costs depend on model usage as well as indexing and search overhead; requests alone do not describe token spend. That is relevant to forecast end-to-end AI workload charges, but the present scenario requires an ability to diagnose serving quota exhaustion under bursts.

Question 248

A product page has informative diagrams that change each release. What QA belongs in the image pipeline?

  1. Reuse the original release’s alt descriptions without checking whether updated diagrams changed.
  2. Use an empty decorative alt attribute rather than describing irrelevant decoration
  3. Review generated alternatives against the final published image versions
  4. Reuse the last release's generated image descriptions without comparing them against the newly published diagrams.

Explanation

Correct Answer

C:

Text alternatives must correspond to what users actually see after edits or revisions. Here the relevant capability must let the team keep accessibility descriptions aligned with final media. An alternative targeting describe complex diagrams beyond short alt text solves another problem.

Incorrect Answers

A:

Diagram revisions can change labels and relationships. Revalidate visual facts and equivalent descriptions with every relevant release.

B:

Decorative images should not add noise for screen-reader navigation. The function serves to omit noninformative decoration from spoken content, whereas the actual requirement is to keep accessibility descriptions aligned with final media.

D:

An alt description may become misleading when a diagram changes. Validate each alternative against the actual final image and surrounding content.

Question 249

A meeting application must turn spoken English into searchable text. Which audio capability fits?

  1. Save only speaker labels from the meeting recording and discard all spoken words before search indexing.
  2. Use speech-to-text transcription on the recorded audio
  3. Use phrase hints or a supported speech customization with domain vocabulary
  4. Consume streaming speech-recognition events and display partial and final hypotheses

Explanation

Correct Answer

B:

Automatic speech recognition transforms speech into timed or textual words that can be indexed. Here the relevant capability must let the team transcribe spoken audio into text. An alternative targeting retain who spoke each transcript segment solves another problem.

Incorrect Answers

A:

Speaker metadata can attribute speech but cannot make the actual utterances searchable. Speech recognition must produce transcript text, with timing and speaker metadata retained where useful.

C:

Domain vocabulary can reduce recognition errors for uncommon names and part numbers. That capability is appropriate for improve recognition of specialist spoken terms, not this separate task of finding how to transcribe spoken audio into text.

D:

Streaming ASR provides incremental captions without waiting for the entire recording. Here the needed outcome is to transcribe spoken audio into text; an option designed to generate low-latency live captions would not directly provide it.

Question 250

A product manual contains short procedures separated by headings. What chunking design is most useful?

  1. Chunk by coherent sections while retaining headings and parent document IDs
  2. Cut every procedure into fixed 100-character windows without preserving headings, parent document references or section boundaries.
  3. Preserve page and source offsets alongside OCR-derived chunks
  4. Re-embed updated chunks and replace their indexed vector records

Explanation

Correct Answer

A:

Semantic chunks improve retrieval relevance and preserve the surrounding procedure context. This resolves the specific need to split manuals on meaningful section boundaries, while a measure that helps retain nearby definitions across chunk boundaries addresses a different requirement.

Incorrect Answers

B:

Character-only windows can split a step from its prerequisites and lose the surrounding procedure heading. Semantically coherent sections and provenance are necessary for useful manual retrieval.

C:

Page-level lineage allows precise citations and verification of extracted claims. This helps when a team needs to enable precise citations to scanned pages; it does not establish how to split manuals on meaningful section boundaries.

D:

Stale embeddings represent the prior content and degrade semantic retrieval after edits. The function serves to refresh vectors when source chunks change, whereas the actual requirement is to split manuals on meaningful section boundaries.

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