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 26

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.

Question 251

A customer-facing agent needs to remember verified approvals during one task, keep concurrent conversations isolated and avoid retaining secrets. Which THREE memory rules are appropriate? Choose THREE.

  1. Minimize conversation memory and redact secrets before persistence or replay
  2. Put every customer's approval and conversation transcript in one shared permanent memory record.
  3. Maintain a short ordered conversation and approval state tied to the task
  4. Keep every historical conversation and credential indefinitely for convenient replay.
  5. Ask for missing required input before invoking account-changing tools
  6. Tag each conversation and tool response with stable isolated session and task identifiers

Explanation

Correct Answers

A:

Retaining only necessary state reduces exposure of sensitive credentials in future prompts. Task state, secret minimization and stable session isolation address distinct risks.

C:

Relevant conversation state keeps authorization and progress coherent across tool cycles.

F:

Correlation IDs prevent results and memory from leaking across concurrent requests.

Incorrect Answers

B:

Cross-user shared state creates confidentiality and authorization failures. Maintain only necessary task approvals under isolated session and user scopes, with secret redaction.

D:

Indefinite retention of credentials and stale interactions conflicts with the explicit minimization and session-safety requirements.

E:

Clarifying mandatory parameters prevents incorrect side effects based on invented assumptions. Its suitable role is to collect missing mandatory task context; it would leave the specific need to preserve task approvals while isolating sessions and redacting secrets unresolved.

Question 252

Clients frequently receive HTTP 429 during peak inference load even though their API keys are valid. Which response strategy is safe?

  1. Retry every throttled request immediately with no delay until it succeeds.
  2. Respect Retry-After, bound exponential backoff, and adjust serving quotas or workload concurrency where justified.
  3. Treat 429 as a permanent authentication failure and rotate the API key.
  4. Open more client connections to bypass the provider's tokens-per-minute limit.

Explanation

Correct Answer

B:

Rate limiting indicates excess allowed request or token throughput, not necessarily invalid credentials; uncontrolled retries can worsen the overload.

Incorrect Answers

A:

Unbounded synchronized retries exacerbate 429 load and can exhaust client resources.

C:

Rate limiting is not proof that credentials are invalid; rotating keys does not raise allowed throughput.

D:

Increased concurrency can worsen throttling without changing the deployed quota.

Question 253

A learning platform publishes a complex chart and a decorative border image. Which TWO accessibility choices convey the right information without noise? Choose TWO.

  1. Use an empty decorative alt attribute rather than describing irrelevant decoration
  2. Approve generated alternative text based on visual layout alone without verifying that the chart’s values and trend are conveyed.
  3. Narrate the decorative border in detail but omit a data-equivalent explanation for the complex chart.
  4. Provide a concise alt summary and a separate detailed equivalent description
  5. Describe every decorative border pixel in lengthy alt text while reducing the informative chart to the phrase “chart image”.

Explanation

Correct Answers

A:

Decorative images should not add noise for screen-reader navigation. Empty alternative text for decoration and a detailed text equivalent for the complex chart are complementary accessibility practices.

D:

Complex instructional graphics often need a longer structured text alternative.

Incorrect Answers

B:

A visual review that never checks the accessible equivalent can miss the chart’s critical trend and data. The required choices distinguish decorative content from meaningful chart explanation.

C:

Decoration should not add screen-reader noise, while an information-rich chart requires an adequate text alternative or adjacent detailed equivalent.

E:

Verbose decoration adds noise while omitting meaningful chart relationships. The border should be decorative and the chart should have an accessible summary and detailed equivalent.

Question 254

A call-center system receives stereo audio with the customer on one channel and agent on another. What metadata should it preserve?

  1. Combine both stereo channels into one speaker before transcription without preserving channel metadata
  2. Use phrase hints or a supported speech customization with domain vocabulary
  3. Use speech-to-text transcription on the recorded audio
  4. Maintain speaker/channel attribution while transcribing both sides

Explanation

Correct Answer

D:

Speaker attribution prevents mixing who made which promise in the resulting transcript. This directly supports the need to retain who spoke each transcript segment, unlike a mechanism primarily intended to generate low-latency live captions.

Incorrect Answers

A:

Collapsing channels removes useful customer/agent attribution. Supported separate-channel or diarization processing better preserves whose speech belongs to which role.

B:

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 retain who spoke each transcript segment.

C:

Automatic speech recognition transforms speech into timed or textual words that can be indexed. Here the needed outcome is to retain who spoke each transcript segment; an option designed to transcribe spoken audio into text would not directly provide it.

Question 255

A chunked policy corpus returns isolated sentences lacking their definitions. What chunk strategy may help?

  1. Remove surrounding headings and definitions from every policy chunk to reduce token use
  2. Use bounded overlap or parent-context retrieval for dependent passages
  3. Split each policy into independent single-sentence chunks and drop enclosing headings, definitions and parent-document context.
  4. Preserve page and source offsets alongside OCR-derived chunks

Explanation

Correct Answer

B:

Adjacent context can preserve critical definitions, although excessive overlap creates duplicates. The practical requirement is to retain nearby definitions across chunk boundaries; a solution built to refresh vectors when source chunks change would not meet that same need.

Incorrect Answers

A:

A short isolated sentence may depend on nearby definitions. Retaining necessary hierarchical context can improve relevance without forcing whole-document retrieval.

C:

Single-sentence fragments reproduce the missing-definition failure. Bounded overlap or parent-context retrieval reconnects dependent clauses with their definitions.

D:

Page-level lineage allows precise citations and verification of extracted claims. That capability is appropriate for enable precise citations to scanned pages, not this separate task of finding how to retain nearby definitions across chunk boundaries.

Question 256

A model calls a payment function with a string for an amount field that requires a nonnegative number. What should the controller do?

  1. Validate submitted arguments against the function schema and reject the invalid call
  2. Define strict identifier fields and resolve identities server-side before writing
  3. Report authorization denied for the payment function even though the supplied amount fails the required numeric input schema.
  4. Expose only explicitly allowed callable functions in the runtime tool registry

Explanation

Correct Answer

A:

Schema validation prevents nonsensical payloads reaching financial side effects. The decisive difference is whether the design can reject tool argument type mismatches; the competing function serves to bind tool operations to verified identity identifiers.

Incorrect Answers

B:

Unambiguous IDs and server checks prevent display-name collisions from changing the wrong record. Here the needed outcome is to reject tool argument type mismatches; an option designed to bind tool operations to verified identity identifiers would not directly provide it.

C:

A malformed amount is a validation failure, not evidence of missing permissions. Reject invalid types and ranges before invoking a payment tool or issuing a misleading authorization result.

D:

A textual description must not grant capabilities outside server-controlled tool permissions. That is relevant to restrict executable tools with an allowlist, but the present scenario requires an ability to reject tool argument type mismatches.

Question 257

Management wants to estimate monthly cost of a proposed retrieval-augmented chatbot. Which usage measurements matter most?

  1. Estimate monthly cost from the maximum output-token setting only, excluding input, retrieval and tool-related usage.
  2. Inspect model deployment throughput limits and subscription quota consumption
  3. Estimate input tokens, output tokens, retrieval calls and forecast traffic volume
  4. Measure only the average latency of the model and retrieval tools, then infer monthly token charges without measuring any requests or tokens.

Explanation

Correct Answer

C:

AI costs depend on model usage as well as indexing and search overhead; requests alone do not describe token spend. In this scenario the team must forecast end-to-end AI workload charges. A feature used to bound output token usage does not substitute for it.

Incorrect Answers

A:

An output cap by itself is not actual consumption. A RAG cost estimate needs measured or projected input and generated tokens, request volume and the relevant retrieval/indexing and tool charges.

B:

Request-per-minute and token-per-minute limits can throttle bursts independently of underlying application server capacity. Its primary purpose is to diagnose serving quota exhaustion under bursts; the stated constraint instead demands a way to forecast end-to-end AI workload charges.

D:

Latency can identify performance bottlenecks but cannot determine billed input or output tokens and search usage. A cost forecast requires request volume, token distributions, retrieval charges and the applicable rate card.

Question 258

An inspection workflow needs to recover distinct components and positions from an equipment photo. What Content Understanding capability fits?

  1. Return one overall image color label and discard the coordinates of the distinct equipment components.
  2. Configure a visual analyzer to identify relevant objects and regions
  3. Evaluate pro-mode Content Understanding for a multi-stage analysis flow
  4. Use a single-task Content Understanding analyzer configured for the required fields

Explanation

Correct Answer

B:

Visual extraction can return object-level evidence for downstream validation or action. The decisive difference is whether the design can extract object and region information from an image; the competing function serves to detect missed visual labels quantitatively.

Incorrect Answers

A:

The inspection needs object-level visual regions. A global color label cannot identify each component or where it appears in the photograph.

C:

Advanced workflows may need a more capable orchestration mode than one fixed extraction. Its primary purpose is to run a richer multimodal understanding pipeline; the stated constraint instead demands a way to extract object and region information from an image.

D:

A focused extraction task can use a single-task pipeline without broad open-ended analysis. That is relevant to perform one narrowly scoped extraction task, but the present scenario requires an ability to extract object and region information from an image.

Question 259

A chatbot returns a summary that includes customer bank account numbers despite a privacy policy. What should be changed?

  1. Apply an output privacy filter and test that sensitive fields are masked in summaries.
  2. Rely on a system prompt saying that private account numbers must never appear.
  3. Disable the source database's encryption to make the text easier to inspect.
  4. Remove all auditing so no generated account values are recorded.

Explanation

Correct Answer

A:

Sensitive data may reappear during generation even after ingestion controls; output controls provide a second protection layer.

Incorrect Answers

B:

A prompt instruction alone does not reliably redact a generated summary or provide a release check.

C:

Removing encryption weakens privacy and does not prevent account numbers in model output.

D:

Suppressing audit records does not sanitize the user-visible answer or prevent disclosure.

Question 260

An extractor returns a numerical value with low confidence in a critical claim form, but the workflow automatically approves payment. What decision gate fits?

  1. Approve the payment because the extractor returned a numeric field.
  2. Route low-confidence critical fields to verification or human review before the financial action.
  3. Raise the payment limit to compensate for possible OCR errors.
  4. Ignore the confidence signal whenever the customer submitted a readable PDF.

Explanation

Correct Answer

B:

A probabilistic extraction should not be the sole authority for a high-impact monetary approval.

Incorrect Answers

A:

A plausible numeric value with low confidence is not validated evidence for a financial decision.

C:

A larger payment ceiling increases risk and does not correct a questionable field.

D:

Page appearance does not verify the extracted critical value or its sign.

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