AWS Certified Generative AI Developer Professional (AIP-C01)

Exam practice questions

15 / 148 questions

Free preview · Answers and explanations includedPage 1 of 2

Implementation and IntegrationAIP-C01 practice

A troubleshooting agent sometimes repeats the same failing lookup and accumulates cost. It already has a prompt asking it to stop after several attempts. Which TWO controls should the application add? Select TWO.

Select two answersAEnforce a maximum step or token budget and an overall execution deadline in the orchestrator.BTrack repeated tool failures and open a circuit breaker or route to a controlled failure path.CIncrease the model’s context window so it can continue looping longer.DTreat a model-generated statement that the task is nearly finished as a budget override.ERetry every tool call indefinitely because the workflow is automated.

AI Safety, Security, and GovernanceAIP-C01 practice

A team enables model invocation logging to troubleshoot document summaries. Prompts and responses can contain confidential customer text. Which TWO controls are necessary for the logging design? Select TWO.

Select two answersAMake the log bucket public so support engineers can find records easily.BAssume model invocation logs never include customer content.CRestrict and encrypt log destinations and configure retention appropriate to the data policy.DLimit or redact sensitive content where feasible and audit access to the retained logs.ERetain all payloads indefinitely because troubleshooting data has no privacy obligations.

AI Safety, Security, and GovernanceAIP-C01 practice

An educational assistant may explain general personal-finance concepts but must not recommend individual investments. Which guardrail configuration most directly expresses this business-topic restriction?

Select one answerAUse high-strength harmful-content filters without defining the business-specific investment-advice restriction.BUse a custom word filter for common investment terms, including terms needed for allowed educational explanations.CRequire the model to append an educational-use disclaimer while allowing it to generate individualized recommendations.DDefine a denied topic for individualized investment recommendations with a clear description and representative examples, then test allowed educational questions.

Foundation Model Integration, Data Management, and ComplianceAIP-C01 practice

Eight product teams are building Bedrock applications. Reviews repeatedly find inconsistent logging, permissive roles, and missing retry policies. Teams still need different models and retrieval sources. What should the platform team provide?

Select one answerAVersioned infrastructure and SDK components with secure defaults, configurable model adapters, observability, and documented exception review.BA shared administrator role and a single unrestricted Bedrock client for every team.COne fixed prompt and one fixed model that every product must use without configuration.DA deployment checklist that leaves all security and reliability implementation to individual teams.

Testing, Validation, and TroubleshootingAIP-C01 practice

A team compares two Bedrock model configurations for an internal assistant. The first was tested on short requests and the second on difficult long requests. Management wants a defensible choice. Which TWO changes are needed? Select TWO.

Select two answersASelect the configuration with the better aggregate score from the unmatched tests.BEvaluate both configurations on the same representative request set with the same acceptance rubric.CChange the rubric after seeing which model performs better.DDiscard failed requests from cost and latency calculations.ECompare quality together with latency and total inference cost, recording all model and prompt settings.

AI Safety, Security, and GovernanceAIP-C01 practice

An AI application must screen generated text before displaying each released chunk. The security team accepts some added latency but does not accept displaying text first and retracting it later. Which streaming configuration fits?

Select one answerAUse asynchronous guardrail processing and rely on stopping only later chunks after a violation.BDisable output checks once the first chunk is generated.CDisplay the full stream and review it only in a nightly audit.DUse synchronous guardrail stream processing so buffered chunks are evaluated before they are sent to the user.

Operational Efficiency and Optimization for GenAI ApplicationsAIP-C01 practice

A prompt deployment accidentally repeats a long reference section in every request. The team wants an early operational alert before the monthly bill is reviewed. Which monitoring approach fits?

Select one answerAWait for a user to report an inaccurate answer.BMonitor only HTTP success rates because successful requests have normal cost.CMonitor input and output token metrics with suitable baselines and alarms, correlated with deployment and prompt-version identifiers.DAssume CloudTrail automatically classifies every prompt as cost efficient.

Implementation and IntegrationAIP-C01 practice

A team has purchased Bedrock Provisioned Throughput for a supported model. Application calls still use the base model identifier, and the provisioned resource remains idle. What change directs calls to the purchased capacity?

Select one answerASet modelId in the supported inference request to the ARN of the provisioned model resource.BIncrease the number of provisioned model units while leaving the invocation target unchanged.CEnable response streaming while continuing to invoke the base model identifier.DRaise the client retry limit so more on-demand calls are attempted.

Operational Efficiency and Optimization for GenAI ApplicationsAIP-C01 practice

An assistant repeatedly sends a long, unchanged reference document followed by a different question. The selected model supports explicit Bedrock prompt caching. Users still need a newly generated answer to each question. What should the developer configure?

Select one answerAPlace the reusable content in a stable eligible prefix with supported cache checkpoints and append the changing question afterward.BReturn the first generated answer for every subsequent question.CRandomize the reference document order on every request to improve cache reuse.DCache only the short changing question and omit the repeated reference text.

Foundation Model Integration, Data Management, and ComplianceAIP-C01 practice

A rail operator needs an assistant that explains disruption policies and reports current platform changes. Policies are revised weekly, while platform assignments change every minute. Responses must cite the applicable policy and use authorized live operational data. Which architecture meets these requirements?

Select one answerARetrieve versioned policy passages from a knowledge base and call an authorized operational API for live assignments before composing the answer.BFine-tune a model weekly on both policy documents and the previous week of platform assignments.CPut all platform events and policies into the system prompt at the beginning of each user session.DUse a knowledge base refreshed weekly for both policies and live assignments.

Operational Efficiency and Optimization for GenAI ApplicationsAIP-C01 practice

A team wants to cache generated answers for repeated policy questions. Policies change, users have different document access, and prompt versions affect the answer. Which TWO controls prevent unsafe or stale reuse? Select TWO.

Select two answersAInclude the relevant authorization scope and model, prompt, and corpus versions in cache eligibility or keys.BUse only the normalized question text as a global key for every tenant.CReuse answers permanently because the wording of the question is unchanged.DTreat a high semantic similarity score as authorization to expose another user’s answer.EInvalidate or expire entries when their source evidence changes and evaluate whether a cached answer still matches the request.

Testing, Validation, and TroubleshootingAIP-C01 practice

A document assistant loses the final sections of long reports. Logs show that preprocessing truncates input before inference to stay within a token limit. What should the developer investigate and change?

Select one answerASwitch to a more capable model while retaining the preprocessing rule that discards the final report sections.BRaise the output-token cap without reconsidering the input budget.CMeasure the full prompt budget, preserve required sections through task-aware chunking or staged summarization, and validate coverage of the complete report.DTreat the truncated document as complete and remove warnings from the logs.

Testing, Validation, and TroubleshootingAIP-C01 practice

A RAG assistant gives incomplete answers. The team cannot tell whether the needed passages were never retrieved or were ignored by the generator. What should it evaluate first to isolate the problem?

Select one answerAChange the generation temperature repeatedly without inspecting retrieved evidence.BGrade only the final answer and assume all errors come from the model.CRun retrieve-only evaluation with labeled relevant passages and inspect ranking, coverage, and retrieval latency before assessing generation on known-good context.DIncrease the answer length while leaving the retrieval path unobserved.

Implementation and IntegrationAIP-C01 practice

A team has several Lambda functions and OpenAPI-described services that agents should discover through one MCP-compatible endpoint. It wants to minimize custom protocol hosting and centralize authentication. Which service should it evaluate?

Select one answerAA custom ECS-hosted MCP gateway with manually implemented protocol adapters and credential exchange for every target.BAmazon Bedrock AgentCore Gateway with supported targets and configured inbound and outbound authentication.CBedrock agent action groups configured separately for each agent, without a shared MCP endpoint for other frameworks.DSeparate MCP servers for each existing service, each with its own discovery endpoint and authentication implementation.

Foundation Model Integration, Data Management, and ComplianceAIP-C01 practice

A field-maintenance application must interpret equipment photographs, return structured findings, and run in an approved AWS Region. Three candidate Bedrock models have different capabilities and prices. How should the developer select a model?

Select one answerASelect a low-cost text model and require the technician to manually describe each image, despite the image-interpretation requirement.BSelect the highest-scoring general benchmark model before checking deployment constraints.CFirst filter candidates by regional availability and required input/output capabilities, then benchmark eligible models on representative maintenance cases.DSelect the model with the strongest text benchmark and defer image and output-format validation until after deployment.

Complete your exam preparation

Continue with the full question bank

Practise free with answers, explanations, and Community discussions. Join Community to unlock the complete question bank and timed mock exams.

Join Community Choose access for your certification

AWS Certified Generative AI Developer Professional (AIP-C01)

Exam practice questions

15 / 148 questions

Free preview · Answers and explanations includedPage 1 of 2

Implementation and IntegrationAIP-C01 practice

A troubleshooting agent sometimes repeats the same failing lookup and accumulates cost. It already has a prompt asking it to stop after several attempts. Which TWO controls should the application add? Select TWO.

AI Safety, Security, and GovernanceAIP-C01 practice

A team enables model invocation logging to troubleshoot document summaries. Prompts and responses can contain confidential customer text. Which TWO controls are necessary for the logging design? Select TWO.

AI Safety, Security, and GovernanceAIP-C01 practice

An educational assistant may explain general personal-finance concepts but must not recommend individual investments. Which guardrail configuration most directly expresses this business-topic restriction?

Foundation Model Integration, Data Management, and ComplianceAIP-C01 practice

Eight product teams are building Bedrock applications. Reviews repeatedly find inconsistent logging, permissive roles, and missing retry policies. Teams still need different models and retrieval sources. What should the platform team provide?

Testing, Validation, and TroubleshootingAIP-C01 practice

A team compares two Bedrock model configurations for an internal assistant. The first was tested on short requests and the second on difficult long requests. Management wants a defensible choice. Which TWO changes are needed? Select TWO.

AI Safety, Security, and GovernanceAIP-C01 practice

An AI application must screen generated text before displaying each released chunk. The security team accepts some added latency but does not accept displaying text first and retracting it later. Which streaming configuration fits?

Operational Efficiency and Optimization for GenAI ApplicationsAIP-C01 practice

A prompt deployment accidentally repeats a long reference section in every request. The team wants an early operational alert before the monthly bill is reviewed. Which monitoring approach fits?

Implementation and IntegrationAIP-C01 practice

A team has purchased Bedrock Provisioned Throughput for a supported model. Application calls still use the base model identifier, and the provisioned resource remains idle. What change directs calls to the purchased capacity?

Operational Efficiency and Optimization for GenAI ApplicationsAIP-C01 practice

An assistant repeatedly sends a long, unchanged reference document followed by a different question. The selected model supports explicit Bedrock prompt caching. Users still need a newly generated answer to each question. What should the developer configure?

Foundation Model Integration, Data Management, and ComplianceAIP-C01 practice

A rail operator needs an assistant that explains disruption policies and reports current platform changes. Policies are revised weekly, while platform assignments change every minute. Responses must cite the applicable policy and use authorized live operational data. Which architecture meets these requirements?

Operational Efficiency and Optimization for GenAI ApplicationsAIP-C01 practice

A team wants to cache generated answers for repeated policy questions. Policies change, users have different document access, and prompt versions affect the answer. Which TWO controls prevent unsafe or stale reuse? Select TWO.

Testing, Validation, and TroubleshootingAIP-C01 practice

A document assistant loses the final sections of long reports. Logs show that preprocessing truncates input before inference to stay within a token limit. What should the developer investigate and change?

Testing, Validation, and TroubleshootingAIP-C01 practice

A RAG assistant gives incomplete answers. The team cannot tell whether the needed passages were never retrieved or were ignored by the generator. What should it evaluate first to isolate the problem?

Implementation and IntegrationAIP-C01 practice

A team has several Lambda functions and OpenAPI-described services that agents should discover through one MCP-compatible endpoint. It wants to minimize custom protocol hosting and centralize authentication. Which service should it evaluate?

Foundation Model Integration, Data Management, and ComplianceAIP-C01 practice

A field-maintenance application must interpret equipment photographs, return structured findings, and run in an approved AWS Region. Three candidate Bedrock models have different capabilities and prices. How should the developer select a model?

Complete your exam preparation

Continue with the full question bank

Practise free with answers, explanations, and Community discussions. Join Community to unlock the complete question bank and timed mock exams.

Join Community Choose access for your certification