Question 320
At a legal case-management provider, engineers need to keep FM invocation access off public internet paths where supported. The current deployment processes customer-specific information. The team wants a solution that can be validated through targeted test cases. Which change should the team prioritize?
- Use supported VPC interface endpoints and restrictive IAM/network controls
- Assume TLS encryption alone makes every path private
- Expose internal endpoints publicly without authentication
- Embed root account keys in model prompts
Explanation
Correct Answer
A:Private connectivity and identity restrictions address separate network and authorization concerns. In a legal case-management provider, the engineering team should verify this design against its stated objective to keep FM invocation access off public internet paths where supported, including the requirement that the current deployment processes customer-specific information.
Incorrect Answers
B:The suggested action—assume TLS encryption alone makes every path private—does not establish the required control for a legal case-management provider to keep FM invocation access off public internet paths where supported. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a legal case-management provider. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—expose internal endpoints publicly without authentication—does not establish the required control for a legal case-management provider to keep FM invocation access off public internet paths where supported. It replaces a verifiable engineering control with an untested shortcut in the a legal case-management provider deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—embed root account keys in model prompts—does not establish the required control for a legal case-management provider to keep FM invocation access off public internet paths where supported. The change is applied outside the component responsible for the requirement identified by a legal case-management provider. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 321
At a retail bank, engineers need to monitor vector index freshness and ingestion failures. The deployment has both bursty and sustained workloads. The team wants a solution that can be validated through targeted test cases. Which architecture is best supported by the stated constraints?
- Never record deleted-source propagation
- Measure document lag, failed updates and retrieval-health indicators
- Monitor only model output length
- Assume all index updates succeed after receiving HTTP 202
Explanation
Correct Answer
B:Index-health observability complements model invocation metrics. In a retail bank, the engineering team should verify this design against its stated objective to monitor vector index freshness and ingestion failures, including the requirement that the deployment has both bursty and sustained workloads.
Incorrect Answers
A:The suggested action—never record deleted-source propagation—does not establish the required control for a retail bank to monitor vector index freshness and ingestion failures. It replaces a verifiable engineering control with an untested shortcut in the a retail bank deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—monitor only model output length—does not establish the required control for a retail bank to monitor vector index freshness and ingestion failures. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a retail bank. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—assume all index updates succeed after receiving HTTP 202—does not establish the required control for a retail bank to monitor vector index freshness and ingestion failures. It replaces a verifiable engineering control with an untested shortcut in the a retail bank deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 322
At a multilingual contact center, engineers need to support document lineage for generated answers. The next release must be evaluated against the previous baseline. An architecture review has rejected fixes that only conceal the symptoms. Which change should the team prioritize?
- Use anonymous embeddings with no reference to original documents
- Ask the language model to invent document revision IDs
- Store stable source identifiers, revision timestamps and permissions alongside chunks
- Discard all source identifiers after vectorization
Explanation
Correct Answer
C:Chunk metadata supports traceability, filtering and source updates. In a multilingual contact center, the engineering team should verify this design against its stated objective to support document lineage for generated answers, including the requirement that the next release must be evaluated against the previous baseline.
Incorrect Answers
A:The suggested action—use anonymous embeddings with no reference to original documents—does not establish the required control for a multilingual contact center to support document lineage for generated answers. It replaces a verifiable engineering control with an untested shortcut in the a multilingual contact center deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
B:The suggested action—ask the language model to invent document revision IDs—does not establish the required control for a multilingual contact center to support document lineage for generated answers. The change is applied outside the component responsible for the requirement identified by a multilingual contact center. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—discard all source identifiers after vectorization—does not establish the required control for a multilingual contact center to support document lineage for generated answers. It replaces a verifiable engineering control with an untested shortcut in the a multilingual contact center deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 323
At a municipal services portal, engineers need to deploy AI application changes with automated safety regression gates. The team must preserve existing identity and authorization boundaries. The team wants a solution that can be validated through targeted test cases. Which design decision addresses the underlying issue?
- Omit rollback because models are probabilistic
- Test only whether the web server responds with HTTP 200
- Use a CI/CD pipeline with quality tests, approvals and reversible deployments
- Push a changed prompt directly to every production client
Explanation
Correct Answer
C:AI-specific release gates examine output behavior as well as software build success. In a municipal services portal, the engineering team should verify this design against its stated objective to deploy AI application changes with automated safety regression gates, including the requirement that the team must preserve existing identity and authorization boundaries.
Incorrect Answers
A:The suggested action—omit rollback because models are probabilistic—does not establish the required control for a municipal services portal to deploy AI application changes with automated safety regression gates. The change is applied outside the component responsible for the requirement identified by a municipal services portal. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
B:The suggested action—test only whether the web server responds with HTTP 200—does not establish the required control for a municipal services portal to deploy AI application changes with automated safety regression gates. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a municipal services portal. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—push a changed prompt directly to every production client—does not establish the required control for a municipal services portal to deploy AI application changes with automated safety regression gates. The change is applied outside the component responsible for the requirement identified by a municipal services portal. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 324
At a SaaS procurement platform, engineers need to preserve user conversation context safely across stateless invocations. Monitoring should isolate failures to the responsible component. A recent review found that the current design has no reliable safeguard for this case. What is the most appropriate production approach?
- Assume a stateless Lambda invocation retains memory between executions
- Store scoped conversation state with retention rules and replay only relevant history
- Put all users in a single shared memory record
- Store unlimited conversation history in every prompt
Explanation
Correct Answer
B:External state management preserves context without violating user isolation or token budgets. In a SaaS procurement platform, the engineering team should verify this design against its stated objective to preserve user conversation context safely across stateless invocations, including the requirement that monitoring should isolate failures to the responsible component.
Incorrect Answers
A:The suggested action—assume a stateless Lambda invocation retains memory between executions—does not establish the required control for a SaaS procurement platform to preserve user conversation context safely across stateless invocations. It replaces a verifiable engineering control with an untested shortcut in the a SaaS procurement platform deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—put all users in a single shared memory record—does not establish the required control for a SaaS procurement platform to preserve user conversation context safely across stateless invocations. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a SaaS procurement platform. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—store unlimited conversation history in every prompt—does not establish the required control for a SaaS procurement platform to preserve user conversation context safely across stateless invocations. It replaces a verifiable engineering control with an untested shortcut in the a SaaS procurement platform deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 325
At an energy maintenance provider, engineers need to reconstruct which data revision supported an AI-generated answer. The design must distinguish inference behavior from upstream data quality. The initial assessment isolates this issue from unrelated infrastructure changes. What should the lead engineer recommend?
- Depend on the model to recall a deleted source later
- Record source IDs, version metadata and request correlation information
- Keep only the final natural-language answer
- Replace source metadata with a random label after generation
Explanation
Correct Answer
B:Traceability links model outputs to the exact version of authorized evidence. In an energy maintenance provider, the engineering team should verify this design against its stated objective to reconstruct which data revision supported an AI-generated answer, including the requirement that the design must distinguish inference behavior from upstream data quality.
Incorrect Answers
A:The suggested action—depend on the model to recall a deleted source later—does not establish the required control for an energy maintenance provider to reconstruct which data revision supported an AI-generated answer. The change is applied outside the component responsible for the requirement identified by an energy maintenance provider. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—keep only the final natural-language answer—does not establish the required control for an energy maintenance provider to reconstruct which data revision supported an AI-generated answer. It replaces a verifiable engineering control with an untested shortcut in the an energy maintenance provider deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—replace source metadata with a random label after generation—does not establish the required control for an energy maintenance provider to reconstruct which data revision supported an AI-generated answer. The change is applied outside the component responsible for the requirement identified by an energy maintenance provider. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 326
At a pharmaceutical documentation team, engineers need to obtain human approval before an agent executes a financial adjustment. The architecture must avoid unnecessary coupling between components. A recent review found that the current design has no reliable safeguard for this case. What should the lead engineer recommend?
- Let the model infer user consent from tone
- Complete the adjustment first and request approval later
- Store an unverified approval in the prompt
- Suspend the workflow for an explicit approval decision before the write action
Explanation
Correct Answer
D:Human-in-the-loop checkpoints should gate consequential actions before execution. In a pharmaceutical documentation team, the engineering team should verify this design against its stated objective to obtain human approval before an agent executes a financial adjustment, including the requirement that the architecture must avoid unnecessary coupling between components.
Incorrect Answers
A:The suggested action—let the model infer user consent from tone—does not establish the required control for a pharmaceutical documentation team to obtain human approval before an agent executes a financial adjustment. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a pharmaceutical documentation team. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
B:The suggested action—complete the adjustment first and request approval later—does not establish the required control for a pharmaceutical documentation team to obtain human approval before an agent executes a financial adjustment. It replaces a verifiable engineering control with an untested shortcut in the a pharmaceutical documentation team deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—store an unverified approval in the prompt—does not establish the required control for a pharmaceutical documentation team to obtain human approval before an agent executes a financial adjustment. The change is applied outside the component responsible for the requirement identified by a pharmaceutical documentation team. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 327
At a retail bank, engineers need to resolve truncated answers when prompts exceed model context limits. Monitoring should isolate failures to the responsible component. The team wants a solution that can be validated through targeted test cases. What should the lead engineer recommend?
- Change IAM credentials to expand a token limit
- Inspect token budgets and adapt history, retrieval context and output allowances
- Raise temperature to make the context window larger
- Remove HTTP retry logic to increase model memory
Explanation
Correct Answer
B:Context-window errors require input and output token budgeting, not authorization changes. In a retail bank, the engineering team should verify this design against its stated objective to resolve truncated answers when prompts exceed model context limits, including the requirement that monitoring should isolate failures to the responsible component.
Incorrect Answers
A:The suggested action—change IAM credentials to expand a token limit—does not establish the required control for a retail bank to resolve truncated answers when prompts exceed model context limits. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a retail bank. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—raise temperature to make the context window larger—does not establish the required control for a retail bank to resolve truncated answers when prompts exceed model context limits. The change is applied outside the component responsible for the requirement identified by a retail bank. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—remove HTTP retry logic to increase model memory—does not establish the required control for a retail bank to resolve truncated answers when prompts exceed model context limits. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a retail bank. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 328
At a payment operations team, engineers need to change embedding models without corrupting nearest-neighbor comparisons. The implementation must remain maintainable across model upgrades. An architecture review has rejected fixes that only conceal the symptoms. What should be implemented before scaling this workflow?
- Re-embed the corpus and queries into the selected model's compatible vector space
- Compare new query vectors with old embeddings without validation
- Change only the vector database display name
- Increase the vector similarity threshold arbitrarily
Explanation
Correct Answer
A:Embedding migrations require compatible representations and a controlled index cutover. In a payment operations team, the engineering team should verify this design against its stated objective to change embedding models without corrupting nearest-neighbor comparisons, including the requirement that the implementation must remain maintainable across model upgrades.
Incorrect Answers
B:The suggested action—compare new query vectors with old embeddings without validation—does not establish the required control for a payment operations team to change embedding models without corrupting nearest-neighbor comparisons. The change is applied outside the component responsible for the requirement identified by a payment operations team. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—change only the vector database display name—does not establish the required control for a payment operations team to change embedding models without corrupting nearest-neighbor comparisons. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a payment operations team. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—increase the vector similarity threshold arbitrarily—does not establish the required control for a payment operations team to change embedding models without corrupting nearest-neighbor comparisons. It replaces a verifiable engineering control with an untested shortcut in the a payment operations team deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 329
At an airline operations unit, engineers need to serve a customized model with requirements not supported by the managed model catalog. The team must preserve existing identity and authorization boundaries. The initial assessment isolates this issue from unrelated infrastructure changes. What is the most appropriate production approach?
- Assume every custom model is automatically a Bedrock on-demand model
- Use API Gateway as the model inference engine
- Place model weights in a Lambda environment variable
- Evaluate a SageMaker AI inference endpoint and its resource sizing
Explanation
Correct Answer
D:Managed custom inference endpoints support deployment controls for compatible specialized models. In an airline operations unit, the engineering team should verify this design against its stated objective to serve a customized model with requirements not supported by the managed model catalog, including the requirement that the team must preserve existing identity and authorization boundaries.
Incorrect Answers
A:The suggested action—assume every custom model is automatically a Bedrock on-demand model—does not establish the required control for an airline operations unit to serve a customized model with requirements not supported by the managed model catalog. The change is applied outside the component responsible for the requirement identified by an airline operations unit. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
B:The suggested action—use API Gateway as the model inference engine—does not establish the required control for an airline operations unit to serve a customized model with requirements not supported by the managed model catalog. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by an airline operations unit. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—place model weights in a Lambda environment variable—does not establish the required control for an airline operations unit to serve a customized model with requirements not supported by the managed model catalog. It replaces a verifiable engineering control with an untested shortcut in the an airline operations unit deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
Question 330
At a university research portal, engineers need to restrict retrieval to documents belonging to the requesting tenant. Operators need measurable acceptance criteria before rollout. A recent review found that the current design has no reliable safeguard for this case. What is the most appropriate production approach?
- Index all tenants without an authorization boundary
- Attach tenant metadata and enforce server-side authorization filters at retrieval time
- Filter citations only after unauthorized content is sent to the model
- Rely solely on a prompt saying do not reveal other tenants
Explanation
Correct Answer
B:Authorization must constrain retrieved context before it reaches the generation request. In a university research portal, the engineering team should verify this design against its stated objective to restrict retrieval to documents belonging to the requesting tenant, including the requirement that operators need measurable acceptance criteria before rollout.
Incorrect Answers
A:The suggested action—index all tenants without an authorization boundary—does not establish the required control for a university research portal to restrict retrieval to documents belonging to the requesting tenant. It replaces a verifiable engineering control with an untested shortcut in the a university research portal deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
C:The suggested action—filter citations only after unauthorized content is sent to the model—does not establish the required control for a university research portal to restrict retrieval to documents belonging to the requesting tenant. No demonstrated capability in this scenario guarantees that this step resolves the concern raised by a university research portal. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.
D:The suggested action—rely solely on a prompt saying do not reveal other tenants—does not establish the required control for a university research portal to restrict retrieval to documents belonging to the requesting tenant. It replaces a verifiable engineering control with an untested shortcut in the a university research portal deployment. In this deployment, engineers should verify the missing mechanism rather than assume the proposed step satisfies the acceptance conditions.