Microsoft AI-103 Foundry Infrastructure Deployment And CI/CD Practice Test

 

This practice topic covers Foundry Infrastructure Deployment and CI/CD for Microsoft AI-103. Questions are original and aligned to the current Microsoft skills outline. For broader exam preparation, review the Microsoft AI-103 Exam Dumps page.

Question 1

In a production AI solution, the required outcome is separate development and production. Which action should the engineer take to handle resource/project boundary correctly?

1. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
2. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
3. Use separate Foundry project or resource boundaries so development changes cannot alter production deployments or connections.
4. Make the release change manually in the portal after the pipeline finishes.
5. Use a preview-only deployment even though the production policy requires generally available features.

Correct Answer: C

Correct Answer

Answer C is correct because this choice directly implements the required decision for resource/project boundary for separate development and production. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement resource/project boundary for separate development and production; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement resource/project boundary for separate development and production; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement resource/project boundary for separate development and production; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement resource/project boundary for separate development and production; the decisive requirement in the stem therefore remains unresolved.

 

Question 2

In a production AI solution, the following condition occurs: data residency constrains regions. Which action should the engineer take to handle infrastructure topology correctly?

1. Use the same environment-specific connection string in development and production.
2. Place the AI solution and its dependent resources only in regions permitted by the stated data-residency requirement.
3. Use a preview-only deployment even though the production policy requires generally available features.
4. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
5. Make the release change manually in the portal after the pipeline finishes.

Correct Answer: B

Correct Answer

Answer B is correct because this choice directly implements the required decision for infrastructure topology when data residency constrains regions. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement infrastructure topology when data residency constrains regions; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement infrastructure topology when data residency constrains regions; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement infrastructure topology when data residency constrains regions; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement infrastructure topology when data residency constrains regions; the decisive requirement in the stem therefore remains unresolved.

 

Question 3

An Azure AI workload must handle placement of dependent search and storage services. Which implementation is most appropriate?

1. Place search and storage dependencies where the application can reach them through the required network and regional boundaries.
2. Keep every environment in one Foundry project and rely only on naming conventions for separation.
3. Use a preview-only deployment even though the production policy requires generally available features.
4. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
5. Expose dependent services publicly and rely on an application allow list as the primary boundary.

Correct Answer: A

Correct Answer

Answer A is correct because this choice directly implements the required decision for placement of dependent search and storage services. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer B is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement placement of dependent search and storage services; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement placement of dependent search and storage services; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement placement of dependent search and storage services; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement placement of dependent search and storage services; the decisive requirement in the stem therefore remains unresolved.

 

Question 4

In a production AI solution, the following condition occurs: custom agent code must run. Which action should the engineer take to handle compute selection correctly?

1. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
2. Use a preview-only deployment even though the production policy requires generally available features.
3. Use the same environment-specific connection string in development and production.
4. Expose dependent services publicly and rely on an application allow list as the primary boundary.
5. Run custom agent code on compute that supports the required runtime dependencies, scaling model, and network access.

Correct Answer: E

Correct Answer

Answer E is correct because this choice directly implements the required decision for compute selection when custom agent code must run. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement compute selection when custom agent code must run; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement compute selection when custom agent code must run; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement compute selection when custom agent code must run; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement compute selection when custom agent code must run; the decisive requirement in the stem therefore remains unresolved.

 

Question 5

In a production AI solution, the required outcome is private data sources. Which action should the engineer take to handle network dependency design correctly?

1. Expose dependent services publicly and rely on an application allow list as the primary boundary.
2. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
3. Make the release change manually in the portal after the pipeline finishes.
4. Use private connectivity for the required data sources and design DNS and routing for that private path.
5. Promote an unpinned model alias so a future model revision can replace the tested version automatically.

Correct Answer: D

Correct Answer

Answer D is correct because this choice directly implements the required decision for network dependency design for private data sources. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement network dependency design for private data sources; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement network dependency design for private data sources; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement network dependency design for private data sources; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement network dependency design for private data sources; the decisive requirement in the stem therefore remains unresolved.

 

Question 6

In a production AI solution, the following condition occurs: regional failure is in scope. Which action should the engineer take to handle recovery topology correctly?

1. Use a preview-only deployment even though the production policy requires generally available features.
2. Expose dependent services publicly and rely on an application allow list as the primary boundary.
3. Deploy a recovery topology with the required services and dependencies available outside the failed region.
4. Make the release change manually in the portal after the pipeline finishes.
5. Store one long-lived administrator secret in both the deployment pipeline and application configuration.

Correct Answer: C

Correct Answer

Answer C is correct because this choice directly implements the required decision for recovery topology when regional failure is in scope. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement recovery topology when regional failure is in scope; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement recovery topology when regional failure is in scope; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement recovery topology when regional failure is in scope; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement recovery topology when regional failure is in scope; the decisive requirement in the stem therefore remains unresolved.

 

Question 7

In a production AI solution, the required outcome is repeatable resource provisioning. Which action should the engineer take to handle infrastructure-as-code boundary correctly?

1. Use the same environment-specific connection string in development and production.
2. Represent the repeatable resource boundary as versioned infrastructure-as-code so each environment can be rebuilt consistently.
3. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
4. Make the release change manually in the portal after the pipeline finishes.
5. Expose dependent services publicly and rely on an application allow list as the primary boundary.

Correct Answer: B

Correct Answer

Answer B is correct because this choice directly implements the required decision for infrastructure-as-code boundary for repeatable resource provisioning. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement infrastructure-as-code boundary for repeatable resource provisioning; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement infrastructure-as-code boundary for repeatable resource provisioning; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement infrastructure-as-code boundary for repeatable resource provisioning; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement infrastructure-as-code boundary for repeatable resource provisioning; the decisive requirement in the stem therefore remains unresolved.

 

Question 8

In a production AI solution, the required outcome is predictable sustained demand. Which action should the engineer take to handle deployment capacity option correctly?

1. Use a capacity option sized for the predictable sustained load rather than optimizing only for idle periods.
2. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
3. Make the release change manually in the portal after the pipeline finishes.
4. Use the same environment-specific connection string in development and production.
5. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.

Correct Answer: A

Correct Answer

Answer A is correct because this choice directly implements the required decision for deployment capacity option for predictable sustained demand. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer B is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement deployment capacity option for predictable sustained demand; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement deployment capacity option for predictable sustained demand; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement deployment capacity option for predictable sustained demand; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement deployment capacity option for predictable sustained demand; the decisive requirement in the stem therefore remains unresolved.

 

Question 9

During a deployment architecture review, the service must satisfy the specific runtime trade-off described by this requirement without changing unrelated project boundaries. Which deployment approach should the engineer choose?

1. Expose dependent services publicly and rely on an application allow list as the primary boundary.
2. Keep every environment in one Foundry project and rely only on naming conventions for separation.
3. Use a preview-only deployment even though the production policy requires generally available features.
4. Make the release change manually in the portal after the pipeline finishes.
5. Use an elastic consumption-oriented deployment option for traffic that is highly intermittent and difficult to reserve efficiently.

Correct Answer: E

Correct Answer

Answer E is correct because this choice directly implements the required decision for deployment option for highly intermittent traffic. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement deployment option for highly intermittent traffic; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement deployment option for highly intermittent traffic; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement deployment option for highly intermittent traffic; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement deployment option for highly intermittent traffic; the decisive requirement in the stem therefore remains unresolved.

 

Question 10

A runtime design must distinguish managed prompt agent from hosted custom runtime. Which implementation uses the correct capability for the requirement?

1. Use the same environment-specific connection string in development and production.
2. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
3. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
4. Use a hosted custom runtime when the application requires custom runtime dependencies or execution control beyond the managed prompt agent.
5. Keep every environment in one Foundry project and rely only on naming conventions for separation.

Correct Answer: D

Correct Answer

Answer D is correct because this choice directly implements the required decision for managed prompt agent versus hosted custom runtime. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement managed prompt agent versus hosted custom runtime; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement managed prompt agent versus hosted custom runtime; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement managed prompt agent versus hosted custom runtime; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement managed prompt agent versus hosted custom runtime; the decisive requirement in the stem therefore remains unresolved.

 

Question 11

A workload has a specific constraint: under stated residency rules. The implementation must choose correctly between regional and broader deployment. Which option best satisfies the constraint?

1. Keep every environment in one Foundry project and rely only on naming conventions for separation.
2. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
3. Choose the regional deployment that keeps data and processing inside the required residency boundary.
4. Use the same environment-specific connection string in development and production.
5. Store one long-lived administrator secret in both the deployment pipeline and application configuration.

Correct Answer: C

Correct Answer

Answer C is correct because this choice directly implements the required decision for regional versus broader deployment under stated residency rules. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement regional versus broader deployment under stated residency rules; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement regional versus broader deployment under stated residency rules; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement regional versus broader deployment under stated residency rules; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement regional versus broader deployment under stated residency rules; the decisive requirement in the stem therefore remains unresolved.

 

Question 12

In a production AI solution, the following condition occurs: model capability is unavailable in a region. Which action should the engineer take to handle deployment choice correctly?

1. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
2. Choose a supported deployment location or model alternative that provides the required capability without violating regional constraints.
3. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
4. Keep every environment in one Foundry project and rely only on naming conventions for separation.
5. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.

Correct Answer: B

Correct Answer

Answer B is correct because this choice directly implements the required decision for deployment choice when model capability is unavailable in a region. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement deployment choice when model capability is unavailable in a region; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement deployment choice when model capability is unavailable in a region; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement deployment choice when model capability is unavailable in a region; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement deployment choice when model capability is unavailable in a region; the decisive requirement in the stem therefore remains unresolved.

 

Question 13

A workload has a specific constraint: when production policy forbids previews. The implementation must choose correctly between preview and ga deployment. Which option best satisfies the constraint?

1. Choose the generally available deployment because the production policy explicitly forbids preview-only features.
2. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
3. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
4. Make the release change manually in the portal after the pipeline finishes.
5. Expose dependent services publicly and rely on an application allow list as the primary boundary.

Correct Answer: A

Correct Answer

Answer A is correct because this choice directly implements the required decision for preview versus GA deployment when production policy forbids previews. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer B is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement preview versus GA deployment when production policy forbids previews; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement preview versus GA deployment when production policy forbids previews; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement preview versus GA deployment when production policy forbids previews; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement preview versus GA deployment when production policy forbids previews; the decisive requirement in the stem therefore remains unresolved.

 

Question 14

A runtime design must distinguish deployment name from model catalog identifier. Which implementation uses the correct capability for the requirement?

1. Use the same environment-specific connection string in development and production.
2. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
3. Keep every environment in one Foundry project and rely only on naming conventions for separation.
4. Expose dependent services publicly and rely on an application allow list as the primary boundary.
5. Reference the configured deployment name at runtime; the model catalog identifier alone is not the deployed endpoint name.

Correct Answer: E

Correct Answer

Answer E is correct because this choice directly implements the required decision for deployment name versus model catalog identifier. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement deployment name versus model catalog identifier; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement deployment name versus model catalog identifier; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement deployment name versus model catalog identifier; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement deployment name versus model catalog identifier; the decisive requirement in the stem therefore remains unresolved.

 

Question 15

In a production AI solution, the required outcome is reproducible rollout. Which action should the engineer take to handle model version pinning correctly?

1. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
2. Expose dependent services publicly and rely on an application allow list as the primary boundary.
3. Use the same environment-specific connection string in development and production.
4. Pin the tested model version so later redeployments and rollbacks reproduce the version that passed evaluation.
5. Use a preview-only deployment even though the production policy requires generally available features.

Correct Answer: D

Correct Answer

Answer D is correct because this choice directly implements the required decision for model version pinning for reproducible rollout. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement model version pinning for reproducible rollout; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement model version pinning for reproducible rollout; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement model version pinning for reproducible rollout; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement model version pinning for reproducible rollout; the decisive requirement in the stem therefore remains unresolved.

 

Question 16

In a production AI solution, the workload includes missing tool dependencies. Which action should the engineer take to handle agent deployment configuration correctly?

1. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
2. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
3. Configure the agent deployment only after all required tools, connections, and identities are present and authorized.
4. Make the release change manually in the portal after the pipeline finishes.
5. Use a preview-only deployment even though the production policy requires generally available features.

Correct Answer: C

Correct Answer

Answer C is correct because this choice directly implements the required decision for agent deployment configuration with missing tool dependencies. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement agent deployment configuration with missing tool dependencies; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement agent deployment configuration with missing tool dependencies; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement agent deployment configuration with missing tool dependencies; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement agent deployment configuration with missing tool dependencies; the decisive requirement in the stem therefore remains unresolved.

 

Question 17

In a production AI solution, the required outcome is an existing model deployment. Which action should the engineer take to handle endpoint configuration correctly?

1. Make the release change manually in the portal after the pipeline finishes.
2. Configure the client with the endpoint and deployment identifiers that belong to the existing deployed model.
3. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
4. Keep every environment in one Foundry project and rely only on naming conventions for separation.
5. Store one long-lived administrator secret in both the deployment pipeline and application configuration.

Correct Answer: B

Correct Answer

Answer B is correct because this choice directly implements the required decision for endpoint configuration for an existing model deployment. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement endpoint configuration for an existing model deployment; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement endpoint configuration for an existing model deployment; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement endpoint configuration for an existing model deployment; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement endpoint configuration for an existing model deployment; the decisive requirement in the stem therefore remains unresolved.

 

Question 18

In a production AI solution, the following condition occurs: the resource exists but deployment fails. Which action should the engineer take to handle capacity assignment correctly?

1. Assign sufficient model capacity or quota to the deployment after confirming the resource itself already exists.
2. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
3. Make the release change manually in the portal after the pipeline finishes.
4. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
5. Use the same environment-specific connection string in development and production.

Correct Answer: A

Correct Answer

Answer A is correct because this choice directly implements the required decision for capacity assignment when the resource exists but deployment fails. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer B is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement capacity assignment when the resource exists but deployment fails; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement capacity assignment when the resource exists but deployment fails; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement capacity assignment when the resource exists but deployment fails; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement capacity assignment when the resource exists but deployment fails; the decisive requirement in the stem therefore remains unresolved.

 

Question 19

In a production AI solution, the workload includes a rollback requirement. Which action should the engineer take to handle staged model/agent rollout correctly?

1. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
2. Use the same environment-specific connection string in development and production.
3. Keep every environment in one Foundry project and rely only on naming conventions for separation.
4. Promote an unpinned model alias so a future model revision can replace the tested version automatically.
5. Use a staged rollout with a defined rollback path to the last tested model and agent configuration.

Correct Answer: E

Correct Answer

Answer E is correct because this choice directly implements the required decision for staged model/agent rollout with a rollback requirement. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement staged model/agent rollout with a rollback requirement; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement staged model/agent rollout with a rollback requirement; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement staged model/agent rollout with a rollback requirement; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement staged model/agent rollout with a rollback requirement; the decisive requirement in the stem therefore remains unresolved.

 

Question 20

An Azure AI workload must handle promotion of versioned prompts and configuration together. Which implementation is most appropriate?

1. Use the same environment-specific connection string in development and production.
2. Make the release change manually in the portal after the pipeline finishes.
3. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
4. Promote the tested prompt, model settings, and dependent configuration together as one versioned release artifact.
5. Keep every environment in one Foundry project and rely only on naming conventions for separation.

Correct Answer: D

Correct Answer

Answer D is correct because this choice directly implements the required decision for promotion of versioned prompts and configuration together. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use the same environment-specific connection string in development and production. That can address a neighboring concern, but it does not implement promotion of versioned prompts and configuration together; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement promotion of versioned prompts and configuration together; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement promotion of versioned prompts and configuration together; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement promotion of versioned prompts and configuration together; the decisive requirement in the stem therefore remains unresolved.

 

Question 21

An Azure AI workload must handle environment-specific connection injection during deployment. Which implementation is most appropriate?

1. Keep every environment in one Foundry project and rely only on naming conventions for separation.
2. Expose dependent services publicly and rely on an application allow list as the primary boundary.
3. Inject environment-specific connections during deployment so development and production do not share endpoints or secrets.
4. Make the release change manually in the portal after the pipeline finishes.
5. Use a preview-only deployment even though the production policy requires generally available features.

Correct Answer: C

Correct Answer

Answer C is correct because this choice directly implements the required decision for environment-specific connection injection during deployment. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement environment-specific connection injection during deployment; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement environment-specific connection injection during deployment; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement environment-specific connection injection during deployment; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement environment-specific connection injection during deployment; the decisive requirement in the stem therefore remains unresolved.

 

Question 22

In a production AI solution, the following condition occurs: candidate output regresses. Which action should the engineer take to handle quality gate correctly?

1. Keep every environment in one Foundry project and rely only on naming conventions for separation.
2. Block promotion when repeatable evaluation shows the candidate regresses against the release acceptance criteria.
3. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
4. Make the release change manually in the portal after the pipeline finishes.
5. Promote an unpinned model alias so a future model revision can replace the tested version automatically.

Correct Answer: B

Correct Answer

Answer B is correct because this choice directly implements the required decision for quality gate when candidate output regresses. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement quality gate when candidate output regresses; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement quality gate when candidate output regresses; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement quality gate when candidate output regresses; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would promote an unpinned model alias so a future model revision can replace the tested version automatically. That can address a neighboring concern, but it does not implement quality gate when candidate output regresses; the decisive requirement in the stem therefore remains unresolved.

 

Question 23

An Azure AI workload must handle rollback of a failed release using pinned artifacts. Which implementation is most appropriate?

1. Roll back by redeploying the previously tested pinned artifacts instead of reconstructing the prior release manually.
2. Keep every environment in one Foundry project and rely only on naming conventions for separation.
3. Make the release change manually in the portal after the pipeline finishes.
4. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
5. Expose dependent services publicly and rely on an application allow list as the primary boundary.

Correct Answer: A

Correct Answer

Answer A is correct because this choice directly implements the required decision for rollback of a failed release using pinned artifacts. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer B is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement rollback of a failed release using pinned artifacts; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would make the release change manually in the portal after the pipeline finishes. That can address a neighboring concern, but it does not implement rollback of a failed release using pinned artifacts; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement rollback of a failed release using pinned artifacts; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement rollback of a failed release using pinned artifacts; the decisive requirement in the stem therefore remains unresolved.

 

Question 24

In a production AI solution, the required outcome is unattended resource deployment. Which action should the engineer take to handle pipeline credentials correctly?

1. Use a preview-only deployment even though the production policy requires generally available features.
2. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.
3. Keep every environment in one Foundry project and rely only on naming conventions for separation.
4. Expose dependent services publicly and rely on an application allow list as the primary boundary.
5. Use workload identity or federated credentials for unattended deployment instead of embedding a long-lived administrator secret.

Correct Answer: E

Correct Answer

Answer E is correct because this choice directly implements the required decision for pipeline credentials for unattended resource deployment. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would use a preview-only deployment even though the production policy requires generally available features. That can address a neighboring concern, but it does not implement pipeline credentials for unattended resource deployment; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement pipeline credentials for unattended resource deployment; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement pipeline credentials for unattended resource deployment; the decisive requirement in the stem therefore remains unresolved.

Answer D is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement pipeline credentials for unattended resource deployment; the decisive requirement in the stem therefore remains unresolved.

 

Question 25

In a production AI solution, the condition appears after an out-of-band configuration change. Which action should the engineer take to handle deployment drift detection correctly?

1. Expose dependent services publicly and rely on an application allow list as the primary boundary.
2. Store one long-lived administrator secret in both the deployment pipeline and application configuration.
3. Keep every environment in one Foundry project and rely only on naming conventions for separation.
4. Detect and reconcile out-of-band configuration changes against the versioned desired deployment state.
5. Skip rollback artifacts and recreate the prior state from memory if the rollout fails.

Correct Answer: D

Correct Answer

Answer D is correct because this choice directly implements the required decision for deployment drift detection after an out-of-band configuration change. It addresses the scenario at the correct stage of the Azure AI solution.

Incorrect Answers

Answer A is incorrect because this choice would expose dependent services publicly and rely on an application allow list as the primary boundary. That can address a neighboring concern, but it does not implement deployment drift detection after an out-of-band configuration change; the decisive requirement in the stem therefore remains unresolved.

Answer B is incorrect because this choice would store one long-lived administrator secret in both the deployment pipeline and application configuration. That can address a neighboring concern, but it does not implement deployment drift detection after an out-of-band configuration change; the decisive requirement in the stem therefore remains unresolved.

Answer C is incorrect because this choice would keep every environment in one Foundry project and rely only on naming conventions for separation. That can address a neighboring concern, but it does not implement deployment drift detection after an out-of-band configuration change; the decisive requirement in the stem therefore remains unresolved.

Answer E is incorrect because this choice would skip rollback artifacts and recreate the prior state from memory if the rollout fails. That can address a neighboring concern, but it does not implement deployment drift detection after an out-of-band configuration change; the decisive requirement in the stem therefore remains unresolved.