Microsoft AI-300 Practice Test 2
Question 10
As an MLOps engineer reviews a model training or inference pipeline, 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 resolution must preserve quality gates, documented artifacts and rollback options. Which acceptance test shows the AI operations defect is corrected?
- Historical inference replay uses only features available before prediction time and realistic performance survives evaluation.
- Authorized auditors can assess model behavior while unauthorized team members cannot recover individual identifiers.
- The selected version meets validation criteria without depending on apparent fit to the training examples alone.
- Recreating a production deployment resolves the same approved model artifact until an authorized promotion changes the version.
Explanation
Correct Answer
B:Authorized auditors can assess model behavior while unauthorized team members cannot recover individual identifiers. Responsible AI review includes protecting personal data in explanatory artifacts, not only calculating model performance metrics. a direct corrective action is to restrict access, redact or aggregate sensitive examples and publish a privacy-reviewed evaluation artifact for authorized reviewers.
Incorrect Answers
A:Historical inference replay uses only features available before prediction time and realistic performance survives evaluation. That addresses a different configuration issue rather than the main requirement. Verify instead: Authorized auditors can assess model behavior while unauthorized team members cannot recover individual identifiers.
C:The selected version meets validation criteria without depending on apparent fit to the training examples alone. That addresses a different configuration issue rather than the main requirement. Verify instead: Authorized auditors can assess model behavior while unauthorized team members cannot recover individual identifiers.
D:Recreating a production deployment resolves the same approved model artifact until an authorized promotion changes the version. That addresses a different configuration issue rather than the main requirement. Verify instead: Authorized auditors can assess model behavior while unauthorized team members cannot recover individual identifiers.
Question 11
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A production deployment references a mutable latest model alias and picks up a candidate model that bypassed validation when another team registered a new version. The operator must safeguard training reproducibility and serving continuity. Which production-safe remediation should the engineer apply to this AI workflow?
- Bind training input directly to the feature component's typed output to establish the correct pipeline dependency.
- Retain a governed archived or retrievable known-good model and its environment until release validation and rollback windows have passed.
- Implement point-in-time correct feature retrieval and package its temporal lookup logic with the registered model version.
- Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
Explanation
Correct Answer
D:Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates. This targets model registry latest alias points to unapproved experiment. Mutable model aliases simplify discovery but are unsafe production identifiers when deployment reproducibility and review are required. Verify the failure through: Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events. Validate success with: Recreating a production deployment resolves the same approved model artifact until an authorized promotion changes the version.
Incorrect Answers
A:Bind training input directly to the feature component's typed output to establish the correct pipeline dependency. This could address pipeline graph omits feature-build dependency, but it does not solve the specified model registry latest alias points to unapproved experiment failure. The matching remediation is: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
B:Retain a governed archived or retrievable known-good model and its environment until release validation and rollback windows have passed. This could address incident rollback artifact was deleted instead of archived, but it does not solve the specified model registry latest alias points to unapproved experiment failure. The matching remediation is: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
C:Implement point-in-time correct feature retrieval and package its temporal lookup logic with the registered model version. This could address online feature retrieval pulls values from after prediction time, but it does not solve the specified model registry latest alias points to unapproved experiment failure. The matching remediation is: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
Question 12
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A repeated model-training command yields unpredictably different evaluation results because its random seed and dataset split are not fixed or logged across Azure ML runs. The operator must safeguard training reproducibility and serving continuity. What measured outcome validates the solution while preserving the stated constraints?
- Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption.
- A rerun executes the revised transform and its output lineage identifies the current code and dataset rather than a stale cached job.
- Repeat runs show explainable variance with recorded seeds and identical inputs where deterministic execution is supported.
- Workers process the intended distinct data shards and aggregated training metrics match the complete dataset definition.
Explanation
Correct Answer
C:Repeat runs show explainable variance with recorded seeds and identical inputs where deterministic execution is supported. ML model reproducibility depends on data, environment and randomness control; a single identical command string does not fix every stochastic dependency. a direct corrective action is to implement feasible deterministic seeding and record input data split and environment versions for each run with acknowledged nondeterministic GPU constraints.
Incorrect Answers
A:Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption. That addresses a different configuration issue rather than the main requirement. Verify instead: Repeat runs show explainable variance with recorded seeds and identical inputs where deterministic execution is supported.
B:A rerun executes the revised transform and its output lineage identifies the current code and dataset rather than a stale cached job. That addresses a different configuration issue rather than the main requirement. Verify instead: Repeat runs show explainable variance with recorded seeds and identical inputs where deterministic execution is supported.
D:Workers process the intended distinct data shards and aggregated training metrics match the complete dataset definition. That addresses a different configuration issue rather than the main requirement. Verify instead: Repeat runs show explainable variance with recorded seeds and identical inputs where deterministic execution is supported.
Question 13
During a Microsoft Foundry generative AI deployment review, the following condition occurs: An enterprise disables public network access to Foundry but its container app still resolves the model service hostname to a public endpoint, resulting in failed inference calls. The deployment must retain governed identities, endpoint contracts and auditable release history. Select TWO actions: one decisive diagnostic and its corresponding targeted remediation.
- Revoke the exposed key, use managed identity or approved secure Key Vault secret references, and scrub sensitive release configuration from ongoing builds.
- Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests.
- Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
- Define a versioned structured response instruction and supported JSON schema contract with parser-based validation.
- Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
Explanation
Correct Answers
C:Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. This is the most relevant evidence for foundry private endpoint dns resolves to public ip. Private network security requires both endpoint approval and correct DNS resolution; a private endpoint creation event does not change every client's name resolution. The matching controlled remediation is: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
E:Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity. This targets foundry private endpoint dns resolves to public ip. Private network security requires both endpoint approval and correct DNS resolution; a private endpoint creation event does not change every client's name resolution. Verify the failure through: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. Validate success with: Requests from the approved VNet resolve the private address and succeed, while public connectivity remains blocked.
Incorrect Answers
A:Revoke the exposed key, use managed identity or approved secure Key Vault secret references, and scrub sensitive release configuration from ongoing builds. This is associated with foundry infrastructure parameters embed a secret in source repository, not the essential evidence-and-fix pair for foundry private endpoint dns resolves to public ip. First examine: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. Then correct: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
B:Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests. This is associated with agent runtime managed identity cannot read retrieval data, not the essential evidence-and-fix pair for foundry private endpoint dns resolves to public ip. First examine: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. Then correct: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
D:Define a versioned structured response instruction and supported JSON schema contract with parser-based validation. This is associated with prompt omits explicit structured answer schema, not the essential evidence-and-fix pair for foundry private endpoint dns resolves to public ip. First examine: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. Then correct: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
Question 14
While preparing a secure Foundry model or agent release, the following condition occurs: A chat assistant must return low-latency answers but the team selects an offline batch inference architecture that produces responses only after a scheduled job completes. The remediation must meet enterprise access and reliability requirements without a broad bypass. Which TWO complementary choices identify the issue and correct it? Select TWO.
- Choose an appropriate available model and supported prompt chunking strategy that meets the workload's context and cost requirements.
- Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest.
- Evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics.
- Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior.
- Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests.
Explanation
Correct Answers
D:Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior. This is the most relevant evidence for interactive inference deployed on unsuitable batch-style compute. Generative chat workloads ordinarily require online serving semantics; scheduled batch compute cannot satisfy an interactive response deadline. The matching controlled remediation is: Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests.
E:Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests. This targets interactive inference deployed on unsuitable batch-style compute. Generative chat workloads ordinarily require online serving semantics; scheduled batch compute cannot satisfy an interactive response deadline. Verify the failure through: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior. Validate success with: Representative chat queries receive valid answers within the approved latency target without waiting for a batch output artifact.
Incorrect Answers
A:Choose an appropriate available model and supported prompt chunking strategy that meets the workload's context and cost requirements. This is associated with foundation model selected without required context window, not the essential evidence-and-fix pair for interactive inference deployed on unsuitable batch-style compute. First examine: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior. Then correct: Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests.
B:Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest. This is associated with managed foundation model deployment lacks compatible runtime image, not the essential evidence-and-fix pair for interactive inference deployed on unsuitable batch-style compute. First examine: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior. Then correct: Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests.
C:Evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics. This is associated with prompt a/b test uses different model versions in each arm, not the essential evidence-and-fix pair for interactive inference deployed on unsuitable batch-style compute. First examine: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior. Then correct: Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests.
Question 15
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: An Azure ML notebook cell executes successfully after an ad hoc pip install, but the same code submitted as a managed job fails because the registered environment excludes that package. The resolution must preserve quality gates, documented artifacts and rollback options. Which observable postchange result proves the intended remediation succeeded?
- The feature job completes before training consumes its versioned output in repeated orchestrated executions.
- Canary traffic follows the approved fraction and rollback to the old deployment is possible when health thresholds fail.
- Local notebook and managed job both import the same pinned dependency and generate comparable model outputs.
- A reviewer compares both runs and identifies the exact code, data, hyperparameters and metrics used without reconstructing them by hand.
Explanation
Correct Answer
C:Local notebook and managed job both import the same pinned dependency and generate comparable model outputs. Interactive notebook state can diverge from a managed training environment; reproducibility needs explicit environment assets. a direct corrective action is to add the vetted package to a versioned Azure ML environment and submit the notebook's training logic as a reproducible job.
Incorrect Answers
A:The feature job completes before training consumes its versioned output in repeated orchestrated executions. That addresses a different configuration issue rather than the main requirement. Verify instead: Local notebook and managed job both import the same pinned dependency and generate comparable model outputs.
B:Canary traffic follows the approved fraction and rollback to the old deployment is possible when health thresholds fail. That addresses a different configuration issue rather than the main requirement. Verify instead: Local notebook and managed job both import the same pinned dependency and generate comparable model outputs.
D:A reviewer compares both runs and identifies the exact code, data, hyperparameters and metrics used without reconstructing them by hand. That addresses a different configuration issue rather than the main requirement. Verify instead: Local notebook and managed job both import the same pinned dependency and generate comparable model outputs.
Question 16
As an AI operations team checks production agent evaluation evidence, the following condition occurs: An evaluation compares chatbot answers against an automatically rewritten query rather than the original user intent, rewarding responses that miss the real troubleshooting question. The workflow must produce measurable quality, safety and performance evidence. Which observable postchange result proves the intended remediation succeeded?
- The evaluation reveals and measures failures that depend on follow-up context and the selected agent version improves those cases.
- A canary regression is visible independently of stable baseline traffic before expanding rollout.
- The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query.
- Multilingual adversarial probes are handled consistently within the approved safety policy and ordinary requests remain usable.
Explanation
Correct Answer
C:The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query. Query rewriting aids retrieval but should not silently redefine what the user's response needs to accomplish. a direct corrective action is to bind relevance evaluation to the authentic user request and assess query rewriting as a separate retrieval-stage metric.
Incorrect Answers
A:The evaluation reveals and measures failures that depend on follow-up context and the selected agent version improves those cases. That addresses a different configuration issue rather than the main requirement. Verify instead: The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query.
B:A canary regression is visible independently of stable baseline traffic before expanding rollout. That addresses a different configuration issue rather than the main requirement. Verify instead: The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query.
D:Multilingual adversarial probes are handled consistently within the approved safety policy and ordinary requests remain usable. That addresses a different configuration issue rather than the main requirement. Verify instead: The pipeline correctly flags answers that fail the user's original task even when they match an intermediate search query.
Question 17
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A data science department provisions an Azure Machine Learning workspace in a sandbox subscription, while production jobs must inherit regulated production policy and approved storage. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Which targeted Azure AI operations change best addresses the documented problem?
- Declare proper resource references or explicit dependencies and staged network prerequisites in the reviewed Bicep deployment.
- Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances.
- Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
- Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately.
Explanation
Correct Answer
D:Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately. This targets azure ml workspace deployed in the wrong subscription. An Azure Machine Learning workspace inherits Azure resource governance from its containing subscription and resource group; a display name does not determine its security boundary. Verify the failure through: Compare the workspace resource ID, subscription tenant, management group policy assignments, linked resources and team deployment manifest. Validate success with: The production team creates and reads workspace assets under the correct policy scope while sandbox users cannot administer production resources.
Incorrect Answers
A:Declare proper resource references or explicit dependencies and staged network prerequisites in the reviewed Bicep deployment. This could address bicep workspace declaration misses dependent private resources, but it does not solve the specified azure ml workspace deployed in the wrong subscription failure. The matching remediation is: Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately.
B: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 azure ml workspace deployed in the wrong subscription failure. The matching remediation is: Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately.
C:Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management. This could address datastore authentication relies on expiring account key, but it does not solve the specified azure ml workspace deployed in the wrong subscription failure. The matching remediation is: Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately.
Question 18
While preparing a secure Foundry model or agent release, the following condition occurs: A Foundry agent invokes its LLM successfully but receives access denied when querying an Azure AI Search index because the assigned managed identity lacks search data reader rights. The remediation must meet enterprise access and reliability requirements without a broad bypass. Which TWO complementary choices identify the issue and correct it? Select TWO.
- Evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics.
- Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
- Inspect Foundry project and resource regions, model deployment regional availability, enterprise residency constraints and API error codes.
- Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests.
- Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
Explanation
Correct Answers
B:Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration. This targets agent runtime managed identity cannot read retrieval data. Foundry model access and Azure AI Search data-plane authorization are separate permission boundaries for a retrieval-augmented agent. Verify the failure through: Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests. Validate success with: Authorized agent queries retrieve permitted indexed chunks while direct access to unrelated indexes remains denied.
D:Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests. This is the most relevant evidence for agent runtime managed identity cannot read retrieval data. Foundry model access and Azure AI Search data-plane authorization are separate permission boundaries for a retrieval-augmented agent. The matching controlled remediation is: Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
Incorrect Answers
A:Evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics. This is associated with prompt a/b test uses different model versions in each arm, not the essential evidence-and-fix pair for agent runtime managed identity cannot read retrieval data. First examine: Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests. Then correct: Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
C:Inspect Foundry project and resource regions, model deployment regional availability, enterprise residency constraints and API error codes. This is associated with foundry project created in mismatched resource location, not the essential evidence-and-fix pair for agent runtime managed identity cannot read retrieval data. First examine: Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests. Then correct: Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
E:Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration. This is associated with foundry bicep rollout omits model deployment dependencies, not the essential evidence-and-fix pair for agent runtime managed identity cannot read retrieval data. First examine: Inspect runtime managed identity, search service RBAC and index permissions, connector authentication, audit failures and client requests. Then correct: Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
Question 19
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A retail demand model triggers constant drift alerts because its production Christmas data is compared only against a summer reference dataset without accounting for known seasonal regimes. The operator must safeguard training reproducibility and serving continuity. What measured outcome validates the solution while preserving the stated constraints?
- Reports distinguish provisional drift/latency signals from later evidence-based accuracy and recall.
- Normal seasonal variation stops generating false incidents and genuine distribution changes still cross documented alert criteria.
- An injected null-rate increase triggers a targeted alert while valid fluctuations remain within control limits.
- Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution.
Explanation
Correct Answer
B:Normal seasonal variation stops generating false incidents and genuine distribution changes still cross documented alert criteria. Data drift measurement depends on a chosen reference distribution; a statistically different season is not automatically a harmful model change. a direct corrective action is to configure a suitable seasonal baseline or context-aware drift monitor while retaining meaningful anomaly thresholds.
Incorrect Answers
A:Reports distinguish provisional drift/latency signals from later evidence-based accuracy and recall. That addresses a different configuration issue rather than the main requirement. Verify instead: Normal seasonal variation stops generating false incidents and genuine distribution changes still cross documented alert criteria.
C:An injected null-rate increase triggers a targeted alert while valid fluctuations remain within control limits. That addresses a different configuration issue rather than the main requirement. Verify instead: Normal seasonal variation stops generating false incidents and genuine distribution changes still cross documented alert criteria.
D:Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution. That addresses a different configuration issue rather than the main requirement. Verify instead: Normal seasonal variation stops generating false incidents and genuine distribution changes still cross documented alert criteria.
Question 20
As an AI operations team checks production agent evaluation evidence, the following condition occurs: A user-facing agent becomes slow after an LLM deployment change, but the team only records total response time and cannot tell retrieval delay from model generation delay. The workflow must produce measurable quality, safety and performance evidence. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Examine evaluation records, chat history serialization, tool traces, single-turn versus multi-turn settings and representative failure conversations.
- Inspect monitoring sample filters, event statuses, language distribution, evaluator eligibility and collected trace metadata.
- Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags.
- Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles.
Explanation
Correct Answer
D:Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles. This is the most relevant evidence for generation latency spike blamed on retrieval without breakdown. Latency budgets should separate agent orchestration, retrieval and generation; total duration alone cannot locate the bottleneck. The matching controlled remediation is: Instrument stage-level latency and compare generation and retrieval spans before applying a targeted capacity or retrieval change.
Incorrect Answers
A:Examine evaluation records, chat history serialization, tool traces, single-turn versus multi-turn settings and representative failure conversations. This would investigate multi-turn agent evaluation uses isolated last message only rather than the documented generation latency spike blamed on retrieval without breakdown issue. The relevant investigation is: Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles.
B:Inspect monitoring sample filters, event statuses, language distribution, evaluator eligibility and collected trace metadata. This would investigate continuous monitoring has no representative sampling coverage rather than the documented generation latency spike blamed on retrieval without breakdown issue. The relevant investigation is: Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles.
C:Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. This would investigate token cost report omits embedding requests rather than the documented generation latency spike blamed on retrieval without breakdown issue. The relevant investigation is: Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles.