Microsoft AI-300 Practice Test 31
Question 300
During a Microsoft Foundry generative AI deployment review, the following condition occurs: A prototype assistant runs only a few tests per day but reserves a large provisioned-throughput deployment, creating a high recurring bill even when no requests arrive. The deployment must retain governed identities, endpoint contracts and auditable release history. Which acceptance test shows the AI operations defect is corrected?
- A realistic image invoice yields extracted required details without unsupported request content or invented visual observations.
- Repeatable IaC releases create the expected project and model deployment before starting inference smoke tests.
- The managed identity invokes the approved endpoint and a revoked developer key is no longer needed for production inference.
- The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected.
Explanation
Correct Answer
D:The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected. Provisioned throughput may be billed for allocated capacity independent of actual use; deployment type should match utilization patterns. a direct corrective action is to move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review.
Incorrect Answers
A:A realistic image invoice yields extracted required details without unsupported request content or invented visual observations. That addresses a different configuration issue rather than the main requirement. Verify instead: The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected.
B:Repeatable IaC releases create the expected project and model deployment before starting inference smoke tests. That addresses a different configuration issue rather than the main requirement. Verify instead: The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected.
C:The managed identity invokes the approved endpoint and a revoked developer key is no longer needed for production inference. That addresses a different configuration issue rather than the main requirement. Verify instead: The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected.
Question 301
During a Foundry generative AI quality and observability incident, the following condition occurs: A Foundry quality monitor aggregates responses from old and new agent deployments, masking a groundedness regression in the new version. The fix must preserve evaluation validity and avoid unnecessary user-data collection. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Examine continuous monitoring configuration, trace deployment IDs, sampling window, evaluator assignment and per-version quality results.
- Inspect tokens per request, input/output token ratios, concurrency, throttling status and model deployment quota.
- Inspect agent instrumentation, OpenTelemetry trace propagation, tool span durations, error attributes and correlated resource logs.
- Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments.
Explanation
Correct Answer
A:Examine continuous monitoring configuration, trace deployment IDs, sampling window, evaluator assignment and per-version quality results. This is the most relevant evidence for foundry continuous evaluation ignores production deployment version. Aggregate monitoring can hide release-specific failures; quality assessment needs traceable version attribution. The matching controlled remediation is: Segment continuous evaluation by model/agent deployment version and configure a focused regression alert for the candidate.
Incorrect Answers
B:Inspect tokens per request, input/output token ratios, concurrency, throttling status and model deployment quota. This would investigate genai throughput report mixes responses and tokens rather than the documented foundry continuous evaluation ignores production deployment version issue. The relevant investigation is: Examine continuous monitoring configuration, trace deployment IDs, sampling window, evaluator assignment and per-version quality results.
C:Inspect agent instrumentation, OpenTelemetry trace propagation, tool span durations, error attributes and correlated resource logs. This would investigate agent traces omit downstream tool failure reason rather than the documented foundry continuous evaluation ignores production deployment version issue. The relevant investigation is: Examine continuous monitoring configuration, trace deployment IDs, sampling window, evaluator assignment and per-version quality results.
D:Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. This would investigate groundedness metric conflated with response fluency rather than the documented foundry continuous evaluation ignores production deployment version issue. The relevant investigation is: Examine continuous monitoring configuration, trace deployment IDs, sampling window, evaluator assignment and per-version quality results.
Question 302
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A responsible AI dashboard is shared widely with raw personal records and explanations containing identifiers rather than access-controlled aggregate findings. The operator must safeguard training reproducibility and serving continuity. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
- Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings.
- Compare candidate and previous deployment model version, environment digest, scoring code, endpoint configuration and historical release manifest.
- Inspect MLflow model directory, MLmodel flavors, inferred signature, environment requirements and endpoint container startup traceback.
Explanation
Correct Answer
B:Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings. This is the most relevant evidence for model evaluation exposes sensitive examples in explainability report. Responsible AI review includes protecting personal data in explanatory artifacts, not only calculating model performance metrics. The matching controlled remediation is: Restrict access, redact or aggregate sensitive examples and publish a privacy-reviewed evaluation artifact for authorized reviewers.
Incorrect Answers
A:Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. This would investigate responsible ai evaluation omits demographic cohort slices rather than the documented model evaluation exposes sensitive examples in explainability report issue. The relevant investigation is: Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings.
C:Compare candidate and previous deployment model version, environment digest, scoring code, endpoint configuration and historical release manifest. This would investigate rollback cannot recover a previous environment version rather than the documented model evaluation exposes sensitive examples in explainability report issue. The relevant investigation is: Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings.
D:Inspect MLflow model directory, MLmodel flavors, inferred signature, environment requirements and endpoint container startup traceback. This would investigate mlflow registry artifact cannot load on serving target rather than the documented model evaluation exposes sensitive examples in explainability report issue. The relevant investigation is: Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings.
Question 303
While optimizing a production RAG or fine-tuned model service, the following condition occurs: A benefits assistant returns a false eligibility statement because its document chunker splits a policy paragraph away from a nearby exception that changes the rule. The resulting model must meet held-out quality and security criteria. What measured outcome validates the solution while preserving the stated constraints?
- Representative domain queries retrieve relevant passages with aligned embedding provenance and the old index can be rolled back safely.
- Previously missed procedure passages are retrieved and grounded answer quality improves without a large irrelevant context increase.
- Evaluation questions about eligibility include the correct exception and still return concise supported citations.
- The customized model returns the approved structured behavior on held-out scenarios without a spike in schema violations.
Explanation
Correct Answer
C:Evaluation questions about eligibility include the correct exception and still return concise supported citations. Chunk segmentation changes which context is available to the model; cutting a qualification out of its parent rule can produce confidently unsupported answers. a direct corrective action is to tune chunk size and overlap with structure-aware splitting so qualifying rules and exceptions remain retrievable together.
Incorrect Answers
A:Representative domain queries retrieve relevant passages with aligned embedding provenance and the old index can be rolled back safely. That addresses a different configuration issue rather than the main requirement. Verify instead: Evaluation questions about eligibility include the correct exception and still return concise supported citations.
B:Previously missed procedure passages are retrieved and grounded answer quality improves without a large irrelevant context increase. That addresses a different configuration issue rather than the main requirement. Verify instead: Evaluation questions about eligibility include the correct exception and still return concise supported citations.
D:The customized model returns the approved structured behavior on held-out scenarios without a spike in schema violations. That addresses a different configuration issue rather than the main requirement. Verify instead: Evaluation questions about eligibility include the correct exception and still return concise supported citations.
Question 304
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A GitHub deployment workload that only uploads models receives broad subscription Contributor rights, enabling it to modify unrelated application infrastructure. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Which targeted Azure AI operations change best addresses the documented problem?
- Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
- Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
- Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances.
- Link the correct private DNS zone and approved storage private endpoint to the managed network path before enforcing storage isolation.
Explanation
Correct Answer
B:Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant. This targets workspace contributor role granted to untrusted automation. An Azure resource role assignment grants management-plane capabilities according to scope and cannot be justified solely by pipeline convenience. Verify the failure through: Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. Validate success with: Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions.
Incorrect Answers
A:Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability. This could address compute cluster cannot allocate requested gpu quota, but it does not solve the specified workspace contributor role granted to untrusted automation failure. The matching remediation is: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
C:Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances. This could address interactive compute instances remain running while unused, but it does not solve the specified workspace contributor role granted to untrusted automation failure. The matching remediation is: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
D:Link the correct private DNS zone and approved storage private endpoint to the managed network path before enforcing storage isolation. This could address aml private workspace storage access fails from compute, but it does not solve the specified workspace contributor role granted to untrusted automation failure. The matching remediation is: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
Question 305
As an AI operations team checks production agent evaluation evidence, the following condition occurs: A retrieval assistant generates fluent answers citing unrelated documents, yet the release dashboard declares success based solely on readability. The workflow must produce measurable quality, safety and performance evidence. Which TWO complementary choices identify the issue and correct it? Select TWO.
- Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments.
- Version the custom evaluator and recalibrate on a fixed reference set before using its scores for cross-version comparison.
- Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.
- Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples.
- Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
Explanation
Correct Answers
A:Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. This is the most relevant evidence for groundedness metric conflated with response fluency. Readability and factual support are distinct dimensions of GenAI output quality; polished prose is not evidence that retrieved sources justify a claim. The matching controlled remediation is: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
E:Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency. This targets groundedness metric conflated with response fluency. Readability and factual support are distinct dimensions of GenAI output quality; polished prose is not evidence that retrieved sources justify a claim. Verify the failure through: Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. Validate success with: Unsupported claims fail the release gate even when the answers are grammatically smooth.
Incorrect Answers
B:Version the custom evaluator and recalibrate on a fixed reference set before using its scores for cross-version comparison. This is associated with custom evaluator silently changes its scoring rubric, not the essential evidence-and-fix pair for groundedness metric conflated with response fluency. First examine: Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. Then correct: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
C:Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend. This is associated with token cost report omits embedding requests, not the essential evidence-and-fix pair for groundedness metric conflated with response fluency. First examine: Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. Then correct: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
D:Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples. This is associated with relevance evaluator uses wrong intended user query, not the essential evidence-and-fix pair for groundedness metric conflated with response fluency. First examine: Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. Then correct: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
Question 306
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A feature-engineering change is merged directly to the main branch without running schema and leakage tests, and the next scheduled model training uses incompatible features. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Which observable postchange result proves the intended remediation succeeded?
- A breaking feature change is rejected before deployment and approved commits are tied to passed evaluation evidence.
- A reviewed release can deploy production resources, but an unreviewed branch workflow cannot obtain production deployment rights.
- The job consumes the intended dataset version and staging files are not silently mixed into production runs.
- The approved branch deploys using short-lived federated tokens, and old client secrets can be revoked without breaking releases.
Explanation
Correct Answer
A:A breaking feature change is rejected before deployment and approved commits are tied to passed evaluation evidence. Git repositories need enforced review and CI checks; simply storing a notebook in source control does not prevent defective changes. a direct corrective action is to require protected pull requests and automated component tests before changes to production training code are merged.
Incorrect Answers
B:A reviewed release can deploy production resources, but an unreviewed branch workflow cannot obtain production deployment rights. That addresses a different configuration issue rather than the main requirement. Verify instead: A breaking feature change is rejected before deployment and approved commits are tied to passed evaluation evidence.
C:The job consumes the intended dataset version and staging files are not silently mixed into production runs. That addresses a different configuration issue rather than the main requirement. Verify instead: A breaking feature change is rejected before deployment and approved commits are tied to passed evaluation evidence.
D:The approved branch deploys using short-lived federated tokens, and old client secrets can be revoked without breaking releases. That addresses a different configuration issue rather than the main requirement. Verify instead: A breaking feature change is rejected before deployment and approved commits are tied to passed evaluation evidence.
Question 307
As an AI operations team checks production agent evaluation evidence, the following condition occurs: A GenAIOps release pipeline writes evaluation scores to a report but deploys an agent even when groundedness falls below the required threshold. The workflow must produce measurable quality, safety and performance evidence. Which observable postchange result proves the intended remediation succeeded?
- The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query.
- An intentionally degraded candidate fails before production promotion and a passing candidate proceeds with an auditable evaluation receipt.
- Authorized engineers can reproduce request failures without finding credentials or customer PII in log exports.
- An adversarial document cannot cause unapproved transactions while authorized research questions still succeed.
Explanation
Correct Answer
B:An intentionally degraded candidate fails before production promotion and a passing candidate proceeds with an auditable evaluation receipt. Automated evaluation provides protection only when results are connected to enforceable deployment decisions. a direct corrective action is to make required quality/safety criteria explicit CI gates and stop deployment when evaluated scores fail the approved tolerances.
Incorrect Answers
A:The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query. That addresses a different configuration issue rather than the main requirement. Verify instead: An intentionally degraded candidate fails before production promotion and a passing candidate proceeds with an auditable evaluation receipt.
C:Authorized engineers can reproduce request failures without finding credentials or customer PII in log exports. That addresses a different configuration issue rather than the main requirement. Verify instead: An intentionally degraded candidate fails before production promotion and a passing candidate proceeds with an auditable evaluation receipt.
D:An adversarial document cannot cause unapproved transactions while authorized research questions still succeed. That addresses a different configuration issue rather than the main requirement. Verify instead: An intentionally degraded candidate fails before production promotion and a passing candidate proceeds with an auditable evaluation receipt.
Question 308
During a Microsoft Foundry generative AI deployment review, the following condition occurs: An order-status assistant sometimes responds with prose rather than JSON because its system instructions never specify the required fields and output format expected by a downstream parser. The deployment must retain governed identities, endpoint contracts and auditable release history. What measured outcome validates the solution while preserving the stated constraints?
- The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset.
- Representative order queries produce parseable permitted fields and reject or safely handle malformed outputs.
- Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content.
- A repeated script completes provisioning under known approved quota and surfaces capacity constraints before application release.
Explanation
Correct Answer
B:Representative order queries produce parseable permitted fields and reject or safely handle malformed outputs. Prompts are part of an application interface; an implicit expectation that models always emit a parser-compatible object is unreliable. a direct corrective action is to define a versioned structured response instruction and supported JSON schema contract with parser-based validation. This matters here because an order-status assistant sometimes responds with prose rather than JSON because its system instructions never….
Incorrect Answers
A:The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset. That addresses a different configuration issue rather than the main requirement. Verify instead: Representative order queries produce parseable permitted fields and reject or safely handle malformed outputs.
C:Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content. That addresses a different configuration issue rather than the main requirement. Verify instead: Representative order queries produce parseable permitted fields and reject or safely handle malformed outputs.
D:A repeated script completes provisioning under known approved quota and surfaces capacity constraints before application release. That addresses a different configuration issue rather than the main requirement. Verify instead: Representative order queries produce parseable permitted fields and reject or safely handle malformed outputs.
Question 309
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A distributed deep-learning job remains queued because the Azure ML compute cluster requests more regional GPU-family vCPU quota than the subscription currently has. The team must preserve approved access, model provenance and private network constraints. Which investigation best verifies the cause before changing production configuration?
- Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity.
- Examine workspace managed network isolation mode, outbound rules, registry destination, DNS and job network diagnostics.
- Check job identity selection, storage data-plane authorization, private endpoint reachability and storage diagnostic access-denied records.
- Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability.
Explanation
Correct Answer
D:Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability. This is the most relevant evidence for compute cluster cannot allocate requested gpu quota. Azure ML computes draw from Azure regional VM family quota; declaring a cluster does not reserve unlimited GPU instances. The matching controlled remediation is: Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
Incorrect Answers
A:Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. This would investigate workspace contributor role granted to untrusted automation rather than the documented compute cluster cannot allocate requested gpu quota issue. The relevant investigation is: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability.
B:Examine workspace managed network isolation mode, outbound rules, registry destination, DNS and job network diagnostics. This would investigate aml managed network blocks model registry egress rather than the documented compute cluster cannot allocate requested gpu quota issue. The relevant investigation is: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability.
C:Check job identity selection, storage data-plane authorization, private endpoint reachability and storage diagnostic access-denied records. This would investigate job identity lacks blob reader data-plane rights rather than the documented compute cluster cannot allocate requested gpu quota issue. The relevant investigation is: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability.
Question 310
During a Microsoft Foundry generative AI deployment review, the following condition occurs: A developer changes a production agent prompt directly in the Foundry portal without committing it, so a later regression cannot be tied to the release history. The deployment must retain governed identities, endpoint contracts and auditable release history. Which targeted Azure AI operations change best addresses the documented problem?
- Design and test balanced representative prompt examples under fairness and privacy constraints, retaining a versioned evaluation baseline.
- Store reviewed prompt templates and configuration in version control and deploy immutable approved revisions through the release workflow.
- Evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics.
- Separate governed development and production environments with distinct project configuration, least-privilege credentials and deployment gates.
Explanation
Correct Answer
B:Store reviewed prompt templates and configuration in version control and deploy immutable approved revisions through the release workflow. This targets prompt hotfix bypasses git review and loses rollback history. Prompt text is production code-like configuration; untracked edits undermine provenance, collaboration and controlled rollback. Verify the failure through: Inspect deployed prompt content, Git repository history, CI release manifest, approval records and available prompt asset versions. Validate success with: Every production prompt version resolves to a reviewed commit and a failed change can be rolled back without copying an unknown portal draft.
Incorrect Answers
A:Design and test balanced representative prompt examples under fairness and privacy constraints, retaining a versioned evaluation baseline. This could address few-shot prompt examples bias outputs toward one subgroup, but it does not solve the specified prompt hotfix bypasses git review and loses rollback history failure. The matching remediation is: Store reviewed prompt templates and configuration in version control and deploy immutable approved revisions through the release workflow.
C:Evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics. This could address prompt a/b test uses different model versions in each arm, but it does not solve the specified prompt hotfix bypasses git review and loses rollback history failure. The matching remediation is: Store reviewed prompt templates and configuration in version control and deploy immutable approved revisions through the release workflow.
D:Separate governed development and production environments with distinct project configuration, least-privilege credentials and deployment gates. This could address foundry environment mixes development and production agent resources, but it does not solve the specified prompt hotfix bypasses git review and loses rollback history failure. The matching remediation is: Store reviewed prompt templates and configuration in version control and deploy immutable approved revisions through the release workflow.