McAfee-Secured Website

Exam Code: AI-103

Exam Name: Developing AI Apps and Agents on Azure

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Apps and Agents Developer Associate

100% Microsoft AI-103 Guaranteed Success With Testking Training Materials

Get Real AI-103 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

187 Questions & Answers with Testing Engine

"Developing AI Apps and Agents on Azure Exam", also known as AI-103 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-103 Exam Engine. Your AI-103 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-103 Practice Test 12

Question 110

A conversation agent's prompts grow with every turn until context-window errors occur. Tool and search usage remain steady. Which cause should be prioritized?

  1. Bound or summarize excess conversation history.
  2. Raise search service replicas because every context failure is a search throughput issue.
  3. Disable authentication headers to gain enough context tokens.
  4. Increase request concurrency to shrink the individual prompt.

Explanation

Correct Answer

A:

Unbounded memory in each request consumes context tokens even when the latest question is short.

Incorrect Answers

B:

Search capacity does not enlarge a deployed model's context window.

C:

Credentials are not the meaningful source of the excessive historical prompt volume.

D:

Parallelism does not reduce the tokens inside one conversation history.

Question 111

A retained customer preference says delivery is allowed to an old address, but the customer revoked it yesterday. The agent continues suggesting the obsolete address. Which memory defect should be investigated?

  1. Re-evaluate the model output without checking whether its persistent memory was updated after the policy change.
  2. Disable identity checks so another user can overwrite the memory.
  3. Treat all remembered preferences as permanent truths once stored.
  4. Check persistent memory for stale post-revision facts.

Explanation

Correct Answer

D:

Durable memory must be updated when authoritative user preferences change or are revoked.

Incorrect Answers

A:

A new generation pass can repeat the same stale stored fact.

B:

Unrestricted writes create privacy and integrity risks rather than safe revocation handling.

C:

Preferences can change and need lifecycle management rather than indefinite trust.

Question 112

A RAG answer cites an outdated contract clause even though a revised version exists in the repository. What evidence chain should investigators follow?

  1. Verify data lineage and version of the cited source document.
  2. Check only the front-end response formatting.
  3. Treat every discrepancy as a model hallucination before checking document versions and provenance.
  4. Increase top-k without checking document versions.

Explanation

Correct Answer

A:

Source lineage reveals whether old content was retained, indexed incorrectly or selected despite newer evidence.

Incorrect Answers

B:

A layout issue does not explain the revision date of retrieved evidence.

C:

A stale or incorrect retrieved document can be the source of an error even when synthesis faithfully follows it.

D:

A larger set may still return outdated content unless lifecycle/version errors are addressed.

Question 113

A distributed agent request passes through Foundry inference, Azure Functions and a retrieval service. Operators cannot correlate failures across them. Which tooling should be integrated?

  1. Use application traces for source-to-tool-to-model correlation.
  2. Change the text embedding model without reviewing telemetry.
  3. Require users to retype the entire conversation when a tool fails.
  4. Record one server CPU snapshot at midnight and ignore request IDs.

Explanation

Correct Answer

A:

Correlated spans and metrics help relate cross-service behavior to one business request.

Incorrect Answers

B:

Embedding changes do not create distributed correlation IDs.

C:

Manual repetition is not a substitute for operational trace data.

D:

A daily CPU point cannot reconstruct a multi-service execution failure.

Question 114

A new inference deployment consistently returns HTTP 200, but it fabricates missing inventory values. Which interpretation is correct?

  1. Only check request length to decide if inventory values are reliable.
  2. Turn off inventory tool logs because the inference endpoint is healthy.
  3. HTTP 200 guarantees model-generated answers are factually correct.
  4. Measure factual task success independently of HTTP status.

Explanation

Correct Answer

D:

A successful HTTP response proves service availability, not that generated content matches business evidence.

Incorrect Answers

A:

Long or short responses can both contain unsupported claims.

B:

External tool evidence is required to verify live business facts.

C:

Status codes describe request handling rather than semantic truth.

Question 115

A model endpoint returns HTTP 429 after a burst of 200 parallel calls. Which operational cause should be inspected first?

  1. Investigate deployment requests-per-minute and tokens-per-minute throttling and apply controlled backoff.
  2. Rebuild the text search index to clear inference quota.
  3. Reset every user's password because 429 indicates authentication failure.
  4. Increase the model output-token cap to make each request cheaper.

Explanation

Correct Answer

A:

A 429 commonly signals rate or quota limits, not proof of invalid credentials.

Incorrect Answers

B:

Search indexing does not directly lift a deployed model's request rate limit.

C:

Authentication problems are ordinarily signaled by other status codes and identity details.

D:

Larger completions may consume more capacity and worsen throttling.

Question 116

A model request returns HTTP 401, while search retrieval is healthy. Which first diagnostic is most relevant?

  1. Check credentials, token audience and expiry on the 401.
  2. Increase the deployment quota because 401 signals a high request rate.
  3. Tune the semantic ranker to return more passages.
  4. Change the prompt examples without updating the access token.

Explanation

Correct Answer

A:

401 indicates an authentication problem with the invoked service, not a semantic-relevance failure.

Incorrect Answers

B:

Capacity pressure typically produces a different response from invalid authentication.

C:

Retrieval ranking does not repair missing or rejected inference credentials.

D:

Prompt content cannot supply the correct caller identity for a rejected request.

Question 117

A newly assigned service principal role still appears unauthorized in immediately following calls. Which possibility should be tested before rewriting the agent?

  1. Delete the existing project so the new role can propagate to another one.
  2. Assume a missing role can be fixed by changing the user's chat question.
  3. Grant subscription Owner to every caller as the quickest workaround.
  4. Investigate role assignment propagation before changing model code.

Explanation

Correct Answer

D:

Azure RBAC changes can take time to propagate and must apply at the required scope.

Incorrect Answers

A:

Resource deletion does not diagnose the current assignment and adds risk.

B:

The model prompt does not change Azure RBAC evaluation.

C:

A broad permanent role is disproportionate and violates least privilege.

Question 118

Security requires an AI service to be reachable only through private network paths, but public ingress remains enabled. What configuration should be reviewed?

  1. Change the resource display name to InternalOnly without network changes.
  2. Add model instructions telling unauthenticated callers not to use the endpoint.
  3. Rely solely on a private client IP while the endpoint remains open publicly.
  4. Disable unsupported public ingress where the service permits and validate private endpoint, DNS and VNet connectivity.

Explanation

Correct Answer

D:

Private-only access requires both a usable private path and restriction of the public route.

Incorrect Answers

A:

Names are metadata and do not enforce connectivity restrictions.

B:

Natural-language prompts cannot replace service network access controls.

C:

Client location does not remove an independent public service access path.

Question 119

One application team needs model inference access but must not manage deployments or billing. Which permission principle fits?

  1. Grant subscription Owner to every application service identity.
  2. Use the same global administrator account for all applications.
  3. Assign the caller inference/read-only permissions at the minimum supported scope, not deployment management.
  4. Rely on hidden endpoint URLs as the sole security measure.

Explanation

Correct Answer

C:

Inference-only access should not automatically confer unrelated management privileges.

Incorrect Answers

A:

Owner exposes broad resource modification and user-management capabilities.

B:

Shared privileged identity reduces attribution and violates least privilege.

D:

Endpoint secrecy is not a substitute for identity and RBAC enforcement.

Question 120

An automated service rotates API keys poorly and repeatedly breaks inference. Its host supports managed identity. Which change removes the shared-key lifecycle?

  1. Use managed identity to avoid rotating embedded access keys.
  2. Put a permanent key in a protected model prompt.
  3. Disable authentication for the service during key rotation.
  4. Encode the existing API key with a different alphabet.

Explanation

Correct Answer

A:

Workload identity replaces distributed reusable static keys with managed credential issuance and authorization.

Incorrect Answers

B:

Prompt content is not a secure or auditable credential store.

C:

Unrestricted inference access is not an acceptable reliability workaround.

D:

Reversible encoding does not eliminate static credential rotation.

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