Amazon AWS AI Practitioner AIF-C01 AI ML Lifecycle MLOps And Evaluation Metrics Practice Test

 

AIF-C01 skills 1.3 | 30 original questions

This AWS Certified AI Practitioner AIF-C01 practice test focuses on ai ml lifecycle mlops and evaluation metrics through original scenario-based questions aligned to AWS Exam Guide version 1.1 published April 30, 2026. Use the full ExamSnap AIF-C01 collection for broader practice across all five current exam domains. For broader exam preparation, review the Amazon AWS Certified AI Practitioner AIF-C01 Exam Dumps page.

Instructions: Select the best answer for each question. Review the rationale after answering. Each distractor includes a brief explanation of why it is not the strongest fit for the stated scenario.

Question 1

The risk manager at Proseware Services is preparing a recommendation for a customer-support modernization. The recommendation must observe production inputs, outputs, drift, quality, latency, and failures. Which choice is the best match? Budget has been approved for the project, but the team still wants to avoid unnecessary recurring consumption. The team is comparing 9 candidate designs after a 527-day proof of concept.

1. Model monitoring
2. Repeatable pipeline
3. Experiment tracking
4. Scalable ML system
5. Technical-debt management

Correct answer: A

Why: Monitoring identifies when production behavior has changed or degraded. It directly addresses the requirement in this scenario.

Option review:

A: Monitoring identifies when production behavior has changed or degraded. It directly addresses the requirement in this scenario.

B: Repeatable pipelines reduce variance and make ML delivery more reliable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Experiment tracking creates traceability across model-development iterations. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Scalability is an MLOps concern across compute, data, deployment, and monitoring. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: MLOps treats maintainability and accumulated complexity as production risks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Model monitoring – Monitoring identifies when production behavior has changed or degraded.

Question 2

Lucerne Publishing has completed discovery for a agentic workflow trial. Before implementation, the security architect must decide how to track operating cost normalized by active user or served customer. Which choice best satisfies that requirement? The team will validate the result with representative production examples before rollout. The control owner requires evidence from 6 test groups before the 564-day release review.

1. Resolution time
2. Cost per user
3. Return on investment (ROI)
4. Recall
5. Customer lifetime value (CLV)

Correct answer: B

Why: Cost per user is a business/operational metric rather than a model-quality metric. It directly addresses the requirement in this scenario.

Option review:

A: Resolution time can reveal operational efficiency gains or bottlenecks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Cost per user is a business/operational metric rather than a model-quality metric. It directly addresses the requirement in this scenario.

C: ROI evaluates whether the economic return justifies investment. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Recall is useful when false negatives are especially costly. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: CLV can capture longer-term business impact beyond a single interaction. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Cost per user – Cost per user is a business/operational metric rather than a model-quality metric.

Question 3

While planning a contact-center transformation, Lamna Healthcare identifies this requirement: make an approved model available for production inference. Which option should the risk manager prioritize if the goal is to limit exposure of sensitive data? The pilot has representative data, and the team will measure the selected approach against an agreed acceptance threshold. The project has 3 downstream consumers and a monthly review of approximately 601 sampled interactions.

1. Data selection
2. Feedback
3. Deployment
4. Pre-training
5. Model training

Correct answer: C

Why: Deployment exposes the model through a production serving mechanism or batch workflow. It directly addresses the requirement in this scenario.

Option review:

A: Data selection influences model capability, risk, coverage, and legal/compliance posture. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Feedback closes the lifecycle loop and can reveal issues not seen in offline tests. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Deployment exposes the model through a production serving mechanism or batch workflow. It directly addresses the requirement in this scenario.

D: Pre-training establishes the general capabilities of a foundation model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Training optimizes model parameters from examples. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Deployment – Deployment exposes the model through a production serving mechanism or batch workflow.

Question 4

A proof of concept at Contoso Retail exposed a design decision for the security architect: the solution must train a model from organization-controlled data and training code when specialized behavior or ownership is required. Which option most directly solves that problem? The team will document the rationale for auditors and wants the recommendation to be defensible from the scenario facts. The rollout spans 8 application teams, each using the same approved requirement set for the next 638 days.

1. Fine-tuning
2. Pre-training
3. Continuous pre-training
4. Custom-trained model
5. Managed foundation model from a provider

Correct answer: D

Why: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. It directly addresses the requirement in this scenario.

Option review:

A: Fine-tuning modifies model weights to specialize performance. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Pre-training establishes the base capabilities of a foundation model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Continuous pre-training can deepen domain knowledge while retaining a general pretrained starting point. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. It directly addresses the requirement in this scenario.

E: Managed models lower operational burden and can accelerate adoption when their capabilities and governance fit the use case. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Custom-trained model – Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management.

Question 5

Fourth Coffee is documenting the target state for a sales-assistant rollout. The risk manager needs a solution that can run the model on infrastructure the organization controls when custom runtime, networking, or model-hosting requirements demand it. Which option is the strongest fit? The team wants the least complex technically correct choice that satisfies the requirement. The evaluation set contains examples from 5 business workflows and 675 recent production cases.

1. Serverless inference
2. Managed model API
3. Batch inference job
4. Real-time inference
5. Self-hosted model API

Correct answer: E

Why: Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability. It directly addresses the requirement in this scenario.

Option review:

A: Serverless inference automatically provisions serving capacity and is well suited to variable or infrequent traffic when supported. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Batch execution can be cost-effective when immediate online responses are unnecessary. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Real-time inference is designed for low-latency synchronous responses. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability. It directly addresses the requirement in this scenario.

Learning point: Self-hosted model API – Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability.

Question 6

Margie Travel is reviewing a internal search upgrade. The security architect has one primary requirement: use AWS generative-AI assistants for supported workplace or developer productivity scenarios. Which choice best fits the requirement? The workload has passed basic feasibility checks, so the remaining question is which approach best matches the requirement. The initial rollout covers 712 internal users across 2 business units.

1. Amazon Q
2. Amazon Quick
3. Kiro
4. SageMaker JumpStart
5. Strands Agents

Correct answer: A

Why: Amazon Q provides generative-AI assistant experiences designed for enterprise and developer use cases. It directly addresses the requirement in this scenario.

Option review:

A: Amazon Q provides generative-AI assistant experiences designed for enterprise and developer use cases. It directly addresses the requirement in this scenario.

B: Amazon Quick is an AI assistant/workspace for business data, analytics, applications, and agentic tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Kiro is an agentic development environment that helps turn prompts and specifications into code, tests, and documentation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: SageMaker JumpStart accelerates model adoption through curated pretrained models and starter solutions. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Strands Agents is an AWS-backed open-source SDK for building agents. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Amazon Q – Amazon Q provides generative-AI assistant experiences designed for enterprise and developer use cases.

Question 7

During a design review for School of Fine Art, the risk manager must record data, code, parameters, and results so model experiments can be compared and reproduced. The team also wants to reduce manual handling. What should the team choose? Stakeholders have ruled out a broad redesign and want the choice that most precisely addresses the stated need. The workload processes about 749 requests during its busiest hour and has a documented fallback path.

1. Model retraining
2. Experiment tracking
3. Model monitoring
4. Scalable ML system
5. Technical-debt management

Correct answer: B

Why: Experiment tracking creates traceability across model-development iterations. It directly addresses the requirement in this scenario.

Option review:

A: Retraining is part of maintaining model relevance over time. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Experiment tracking creates traceability across model-development iterations. It directly addresses the requirement in this scenario.

C: Monitoring identifies when production behavior has changed or degraded. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Scalability is an MLOps concern across compute, data, deployment, and monitoring. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: MLOps treats maintainability and accumulated complexity as production risks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Experiment tracking – Experiment tracking creates traceability across model-development iterations.

Question 8

Northwind Analytics is moving a knowledge-assistant rollout from pilot to production. The key decision is how to measure qualitative or scored user perceptions of usefulness and quality. Which option is the strongest fit if the team wants to use current managed AWS capabilities? Operational ownership is already assigned, so the team is comparing technical fit rather than staffing models. The pilot uses 786 representative records from 4 approved data sources.

1. Recall
2. Efficiency
3. Customer feedback
4. Accuracy
5. Precision

Correct answer: C

Why: Customer feedback complements technical metrics by capturing real user outcomes. It directly addresses the requirement in this scenario.

Option review:

A: Recall is useful when false negatives are especially costly. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Efficiency captures productivity improvements such as reduced handling time. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Customer feedback complements technical metrics by capturing real user outcomes. It directly addresses the requirement in this scenario.

D: Accuracy is total correct predictions divided by all predictions. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Precision is useful when false positives are especially costly. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Customer feedback – Customer feedback complements technical metrics by capturing real user outcomes.

Question 9

A workshop at Litware Financial focuses on a single decision: how to gather the raw examples and signals needed for the ML project. Which option should the risk manager recommend? Existing application interfaces can accommodate any of the listed choices, so functional fit is the deciding factor. The first release supports 9 departments and is reviewed every 823 days.

1. Feedback
2. Data selection
3. Pre-training
4. Data collection
5. Model selection

Correct answer: D

Why: Data collection establishes the source material used by downstream preparation and training. It directly addresses the requirement in this scenario.

Option review:

A: Feedback closes the lifecycle loop and can reveal issues not seen in offline tests. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Data selection influences model capability, risk, coverage, and legal/compliance posture. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Pre-training establishes the general capabilities of a foundation model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Data collection establishes the source material used by downstream preparation and training. It directly addresses the requirement in this scenario.

E: Model selection chooses the best fit before customization or deployment. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Data collection – Data collection establishes the source material used by downstream preparation and training.

Question 10

For the personalization program at A. Datum Research, stakeholders need to consume a provider-hosted pretrained model through a managed service instead of operating model infrastructure. Which concept, service, or technique most directly addresses this goal? Assume the required AWS capabilities are available in the selected Region and normal governance controls are in place. The service has a 860-millisecond internal response target for the affected workflow.

1. Continuous pre-training
2. Open-source pretrained model
3. Custom-trained model
4. Fine-tuning
5. Managed foundation model from a provider

Correct answer: E

Why: Managed models lower operational burden and can accelerate adoption when their capabilities and governance fit the use case. It directly addresses the requirement in this scenario.

Option review:

A: Continuous pre-training can deepen domain knowledge while retaining a general pretrained starting point. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Fine-tuning modifies model weights to specialize performance. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Managed models lower operational burden and can accelerate adoption when their capabilities and governance fit the use case. It directly addresses the requirement in this scenario.

Learning point: Managed foundation model from a provider – Managed models lower operational burden and can accelerate adoption when their capabilities and governance fit the use case.

Question 11

Coho Winery is comparing alternatives for its developer-productivity pilot. The risk manager needs to produce predictions for accumulated data on a schedule instead of maintaining an always-on endpoint. Which option is most appropriate while trying to meet a strict latency target? The review committee wants a direct mapping from the requirement to the chosen capability. The team is comparing 3 candidate designs after a 897-day proof of concept.

1. Batch inference job
2. Batch inference
3. Asynchronous inference
4. Serverless inference
5. Real-time inference

Correct answer: A

Why: Batch execution can be cost-effective when immediate online responses are unnecessary. It directly addresses the requirement in this scenario.

Option review:

A: Batch execution can be cost-effective when immediate online responses are unnecessary. It directly addresses the requirement in this scenario.

B: Batch inference processes many records together and is appropriate when results can be produced on a schedule. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Asynchronous inference decouples request submission from result retrieval for longer-running workloads. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Serverless inference automatically provisions serving capacity and is well suited to variable or infrequent traffic when supported. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Real-time inference is designed for low-latency synchronous responses. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Batch inference job – Batch execution can be cost-effective when immediate online responses are unnecessary.

Question 12

An architecture review at Lucerne Retail has narrowed a fraud-review pilot decision to one requirement: perform managed ML development, training, deployment, and operations. What should the security architect select? The solution will serve multiple internal teams, so the recommendation should be reusable without changing the core requirement. The control owner requires evidence from 8 test groups before the 934-day release review.

1. Amazon Bedrock AgentCore
2. Amazon SageMaker AI
3. Amazon Bedrock
4. Amazon Quick
5. Strands Agents

Correct answer: B

Why: SageMaker AI spans the traditional ML lifecycle and also supports GenAI development capabilities. It directly addresses the requirement in this scenario.

Option review:

A: AgentCore provides managed infrastructure and services for operating agents at scale. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: SageMaker AI spans the traditional ML lifecycle and also supports GenAI development capabilities. It directly addresses the requirement in this scenario.

C: Amazon Bedrock provides managed foundation-model access and GenAI building blocks without requiring customers to manage model-serving infrastructure. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Amazon Quick is an AI assistant/workspace for business data, analytics, applications, and agentic tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Strands Agents is an AWS-backed open-source SDK for building agents. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Amazon SageMaker AI – SageMaker AI spans the traditional ML lifecycle and also supports GenAI development capabilities.

Question 13

The risk manager at Tailspin Toys is preparing a recommendation for a analytics modernization. The recommendation must control hidden dependencies, fragile pipelines, and manual model-maintenance burden. Which choice is the best match? The decision must follow the workload characteristics rather than a preference for the largest model or newest service. The project has 5 downstream consumers and a monthly review of approximately 971 sampled interactions.

1. Model monitoring
2. Scalable ML system
3. Technical-debt management
4. Experiment tracking
5. Repeatable pipeline

Correct answer: C

Why: MLOps treats maintainability and accumulated complexity as production risks. It directly addresses the requirement in this scenario.

Option review:

A: Monitoring identifies when production behavior has changed or degraded. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Scalability is an MLOps concern across compute, data, deployment, and monitoring. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: MLOps treats maintainability and accumulated complexity as production risks. It directly addresses the requirement in this scenario.

D: Experiment tracking creates traceability across model-development iterations. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Repeatable pipelines reduce variance and make ML delivery more reliable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Technical-debt management – MLOps treats maintainability and accumulated complexity as production risks.

Question 14

City Power and Light has completed discovery for a compliance-assistant prototype. Before implementation, the security architect must decide how to measure how many predicted positive cases are actually positive. Which choice best satisfies that requirement? The security baseline is already defined; the decision here concerns the specific capability described in the requirement. The rollout spans 2 application teams, each using the same approved requirement set for the next 48 days.

1. Customer feedback
2. Accuracy
3. Cost per user
4. Precision
5. Average revenue per user (ARPU)

Correct answer: D

Why: Precision is useful when false positives are especially costly. It directly addresses the requirement in this scenario.

Option review:

A: Customer feedback complements technical metrics by capturing real user outcomes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Accuracy is total correct predictions divided by all predictions. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Cost per user is a business/operational metric rather than a model-quality metric. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Precision is useful when false positives are especially costly. It directly addresses the requirement in this scenario.

E: ARPU connects an AI experience to monetization outcomes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Precision – Precision is useful when false positives are especially costly.

Question 15

While planning a forecasting initiative, Consolidated Messenger identifies this requirement: measure model behavior on suitable validation or test data before promotion. Which option should the risk manager prioritize if the goal is to limit exposure of sensitive data? The recommendation must solve the stated requirement without introducing unrelated platform complexity. The evaluation set contains examples from 7 business workflows and 85 recent production cases.

1. Data preprocessing
2. Monitoring and retraining
3. Evaluation
4. Fine-tuning
5. Model evaluation

Correct answer: E

Why: Evaluation assesses whether the model meets technical and business acceptance criteria. It directly addresses the requirement in this scenario.

Option review:

A: Preprocessing converts raw data into a usable and consistent form. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Monitoring detects drift or degradation; retraining refreshes the model when justified. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Evaluation is required to compare models and validate readiness. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Fine-tuning changes model weights to better fit a target behavior or domain. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Evaluation assesses whether the model meets technical and business acceptance criteria. It directly addresses the requirement in this scenario.

Learning point: Model evaluation – Evaluation assesses whether the model meets technical and business acceptance criteria.

Question 16

A proof of concept at Nod Publishers exposed a design decision for the security architect: the solution must start from an existing model whose weights or implementation are available under an appropriate license. Which option most directly solves that problem? The design must remain supportable after launch, but no additional feature is required beyond the stated need. The initial rollout covers 122 internal users across 4 business units.

1. Open-source pretrained model
2. Pre-training
3. Custom-trained model
4. Distillation
5. Fine-tuning

Correct answer: A

Why: An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review. It directly addresses the requirement in this scenario.

Option review:

A: An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review. It directly addresses the requirement in this scenario.

B: Pre-training establishes the base capabilities of a foundation model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Distillation aims to reduce serving cost and latency while retaining much of the teacher capability. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Fine-tuning modifies model weights to specialize performance. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Open-source pretrained model – An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review.

Question 17

Fabrikam Health is documenting the target state for a agentic workflow trial. The risk manager needs a solution that can serve predictions through a provider-managed endpoint to reduce infrastructure operations. Which option is the strongest fit? A short pilot window means the team prefers an approach that can be evaluated with clear success criteria. The workload processes about 159 requests during its busiest hour and has a documented fallback path.

1. Batch inference job
2. Managed model API
3. Serverless inference
4. Self-hosted model API
5. Batch inference

Correct answer: B

Why: A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider. It directly addresses the requirement in this scenario.

Option review:

A: Batch execution can be cost-effective when immediate online responses are unnecessary. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider. It directly addresses the requirement in this scenario.

C: Serverless inference automatically provisions serving capacity and is well suited to variable or infrequent traffic when supported. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Batch inference processes many records together and is appropriate when results can be produced on a schedule. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Managed model API – A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider.

Question 18

Wingtip Logistics is reviewing a contact-center transformation. The security architect has one primary requirement: build GenAI applications with managed access to foundation models and related capabilities. Which choice best fits the requirement? The architecture board will reject a choice that addresses a different problem from the one described. The pilot uses 196 representative records from 6 approved data sources.

1. Strands Agents
2. SageMaker JumpStart
3. Amazon Bedrock
4. Amazon Quick
5. Amazon Q

Correct answer: C

Why: Amazon Bedrock provides managed foundation-model access and GenAI building blocks without requiring customers to manage model-serving infrastructure. It directly addresses the requirement in this scenario.

Option review:

A: Strands Agents is an AWS-backed open-source SDK for building agents. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: SageMaker JumpStart accelerates model adoption through curated pretrained models and starter solutions. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Amazon Bedrock provides managed foundation-model access and GenAI building blocks without requiring customers to manage model-serving infrastructure. It directly addresses the requirement in this scenario.

D: Amazon Quick is an AI assistant/workspace for business data, analytics, applications, and agentic tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Amazon Q provides generative-AI assistant experiences designed for enterprise and developer use cases. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Amazon Bedrock – Amazon Bedrock provides managed foundation-model access and GenAI building blocks without requiring customers to manage model-serving infrastructure.

Question 19

During a design review for Trey Research, the risk manager must design training and inference processes that can handle growth in data, users, and models. The team also wants to reduce manual handling. What should the team choose? Budget has been approved for the project, but the team still wants to avoid unnecessary recurring consumption. The first release supports 3 departments and is reviewed every 233 days.

1. Experiment tracking
2. Technical-debt management
3. Model monitoring
4. Scalable ML system
5. Model retraining

Correct answer: D

Why: Scalability is an MLOps concern across compute, data, deployment, and monitoring. It directly addresses the requirement in this scenario.

Option review:

A: Experiment tracking creates traceability across model-development iterations. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: MLOps treats maintainability and accumulated complexity as production risks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Monitoring identifies when production behavior has changed or degraded. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Scalability is an MLOps concern across compute, data, deployment, and monitoring. It directly addresses the requirement in this scenario.

E: Retraining is part of maintaining model relevance over time. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Scalable ML system – Scalability is an MLOps concern across compute, data, deployment, and monitoring.

Question 20

Bellows College is moving a sales-assistant rollout from pilot to production. The key decision is how to compare business benefit generated by the solution with its total cost. Which option is the strongest fit if the team wants to use current managed AWS capabilities? The team will validate the result with representative production examples before rollout. The service has a 270-millisecond internal response target for the affected workflow.

1. User satisfaction
2. Efficiency
3. Average revenue per user (ARPU)
4. Customer feedback
5. Return on investment (ROI)

Correct answer: E

Why: ROI evaluates whether the economic return justifies investment. It directly addresses the requirement in this scenario.

Option review:

A: User satisfaction captures perceived value that technical metrics may miss. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Efficiency captures productivity improvements such as reduced handling time. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: ARPU connects an AI experience to monetization outcomes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Customer feedback complements technical metrics by capturing real user outcomes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: ROI evaluates whether the economic return justifies investment. It directly addresses the requirement in this scenario.

Learning point: Return on investment (ROI) – ROI evaluates whether the economic return justifies investment.

Question 21

A workshop at Blue Yonder Airlines focuses on a single decision: how to fit model parameters using the prepared training data. Which option should the risk manager recommend? The pilot has representative data, and the team will measure the selected approach against an agreed acceptance threshold. The team is comparing 5 candidate designs after a 307-day proof of concept.

1. Model training
2. Feedback
3. Fine-tuning
4. Evaluation
5. Deployment

Correct answer: A

Why: Training optimizes model parameters from examples. It directly addresses the requirement in this scenario.

Option review:

A: Training optimizes model parameters from examples. It directly addresses the requirement in this scenario.

B: Feedback closes the lifecycle loop and can reveal issues not seen in offline tests. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Fine-tuning changes model weights to better fit a target behavior or domain. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Evaluation is required to compare models and validate readiness. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Deployment exposes the model through a production serving mechanism or batch workflow. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Model training – Training optimizes model parameters from examples.

Question 22

For the document-intelligence project at Woodgrove Bank, stakeholders need to train a model from organization-controlled data and training code when specialized behavior or ownership is required. Which concept, service, or technique most directly addresses this goal? The team will document the rationale for auditors and wants the recommendation to be defensible from the scenario facts. The control owner requires evidence from 2 test groups before the 344-day release review.

1. Continuous pre-training
2. Custom-trained model
3. Open-source pretrained model
4. Managed foundation model from a provider
5. Distillation

Correct answer: B

Why: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. It directly addresses the requirement in this scenario.

Option review:

A: Continuous pre-training can deepen domain knowledge while retaining a general pretrained starting point. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. It directly addresses the requirement in this scenario.

C: An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Managed models lower operational burden and can accelerate adoption when their capabilities and governance fit the use case. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Distillation aims to reduce serving cost and latency while retaining much of the teacher capability. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Custom-trained model – Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management.

Question 23

Wide World Importers is comparing alternatives for its knowledge-assistant rollout. The risk manager needs to run the model on infrastructure the organization controls when custom runtime, networking, or model-hosting requirements demand it. Which option is most appropriate while trying to meet a strict latency target? The team wants the least complex technically correct choice that satisfies the requirement. The project has 7 downstream consumers and a monthly review of approximately 381 sampled interactions.

1. Batch inference
2. Managed model API
3. Self-hosted model API
4. Asynchronous inference
5. Serverless inference

Correct answer: C

Why: Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability. It directly addresses the requirement in this scenario.

Option review:

A: Batch inference processes many records together and is appropriate when results can be produced on a schedule. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability. It directly addresses the requirement in this scenario.

D: Asynchronous inference decouples request submission from result retrieval for longer-running workloads. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Serverless inference automatically provisions serving capacity and is well suited to variable or infrequent traffic when supported. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Self-hosted model API – Self-hosting provides more infrastructure control but adds responsibility for scaling, patching, monitoring, and availability.

Question 24

An architecture review at VanArsdel Media has narrowed a claims-processing redesign decision to one requirement: use an agentic IDE and CLI for specification-driven software development. What should the security architect select? The workload has passed basic feasibility checks, so the remaining question is which approach best matches the requirement. The rollout spans 4 application teams, each using the same approved requirement set for the next 418 days.

1. Amazon Q
2. Amazon SageMaker AI
3. Amazon Bedrock
4. Kiro
5. Amazon Bedrock AgentCore

Correct answer: D

Why: Kiro is an agentic development environment that helps turn prompts and specifications into code, tests, and documentation. It directly addresses the requirement in this scenario.

Option review:

A: Amazon Q provides generative-AI assistant experiences designed for enterprise and developer use cases. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: SageMaker AI spans the traditional ML lifecycle and also supports GenAI development capabilities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Amazon Bedrock provides managed foundation-model access and GenAI building blocks without requiring customers to manage model-serving infrastructure. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Kiro is an agentic development environment that helps turn prompts and specifications into code, tests, and documentation. It directly addresses the requirement in this scenario.

E: AgentCore provides managed infrastructure and services for operating agents at scale. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Kiro – Kiro is an agentic development environment that helps turn prompts and specifications into code, tests, and documentation.

Question 25

The risk manager at Datum Dynamics is preparing a recommendation for a personalization program. The recommendation must automate consistent build, test, and deployment steps instead of relying on ad hoc manual work. Which choice is the best match? Stakeholders have ruled out a broad redesign and want the choice that most precisely addresses the stated need. The evaluation set contains examples from 9 business workflows and 455 recent production cases.

1. Model retraining
2. Experiment tracking
3. Model monitoring
4. Technical-debt management
5. Repeatable pipeline

Correct answer: E

Why: Repeatable pipelines reduce variance and make ML delivery more reliable. It directly addresses the requirement in this scenario.

Option review:

A: Retraining is part of maintaining model relevance over time. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Experiment tracking creates traceability across model-development iterations. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Monitoring identifies when production behavior has changed or degraded. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: MLOps treats maintainability and accumulated complexity as production risks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Repeatable pipelines reduce variance and make ML delivery more reliable. It directly addresses the requirement in this scenario.

Learning point: Repeatable pipeline – Repeatable pipelines reduce variance and make ML delivery more reliable.

Question 26

Alpine Ski House has completed discovery for a developer-productivity pilot. Before implementation, the security architect must decide how to balance precision and recall with their harmonic mean. Which choice best satisfies that requirement? Operational ownership is already assigned, so the team is comparing technical fit rather than staffing models. The initial rollout covers 492 internal users across 6 business units.

1. F1 score
2. Recall
3. Resolution time
4. Cross-domain performance
5. Conversion rate

Correct answer: A

Why: F1 summarizes precision and recall when both types of error matter. It directly addresses the requirement in this scenario.

Option review:

A: F1 summarizes precision and recall when both types of error matter. It directly addresses the requirement in this scenario.

B: Recall is useful when false negatives are especially costly. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Resolution time can reveal operational efficiency gains or bottlenecks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Cross-domain testing checks whether broad capability actually transfers to the required business contexts. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Conversion rate is relevant when the application aims to influence transactions or signups. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: F1 score – F1 summarizes precision and recall when both types of error matter.

Question 27

While planning a fraud-review pilot, Humongous Insurance identifies this requirement: clean, transform, and prepare raw data before training or inference. Which option should the risk manager prioritize if the goal is to limit exposure of sensitive data? Existing application interfaces can accommodate any of the listed choices, so functional fit is the deciding factor. The workload processes about 529 requests during its busiest hour and has a documented fallback path.

1. Data selection
2. Data preprocessing
3. Model evaluation
4. Evaluation
5. Deployment

Correct answer: B

Why: Preprocessing converts raw data into a usable and consistent form. It directly addresses the requirement in this scenario.

Option review:

A: Data selection influences model capability, risk, coverage, and legal/compliance posture. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Preprocessing converts raw data into a usable and consistent form. It directly addresses the requirement in this scenario.

C: Evaluation assesses whether the model meets technical and business acceptance criteria. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Evaluation is required to compare models and validate readiness. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Deployment exposes the model through a production serving mechanism or batch workflow. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Data preprocessing – Preprocessing converts raw data into a usable and consistent form.

Question 28

A proof of concept at Graphic Design Institute exposed a design decision for the security architect: the solution must start from an existing model whose weights or implementation are available under an appropriate license. Which option most directly solves that problem? Assume the required AWS capabilities are available in the selected Region and normal governance controls are in place. The pilot uses 566 representative records from 8 approved data sources.

1. Distillation
2. Pre-training
3. Open-source pretrained model
4. Custom-trained model
5. Continuous pre-training

Correct answer: C

Why: An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review. It directly addresses the requirement in this scenario.

Option review:

A: Distillation aims to reduce serving cost and latency while retaining much of the teacher capability. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Pre-training establishes the base capabilities of a foundation model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review. It directly addresses the requirement in this scenario.

D: Custom training offers greater control but usually requires more data, compute, expertise, and lifecycle management. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Continuous pre-training can deepen domain knowledge while retaining a general pretrained starting point. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Open-source pretrained model – An open-source pretrained model can reduce initial training effort but still requires license, security, and operational review.

Question 29

Relecloud is documenting the target state for a compliance-assistant prototype. The risk manager needs a solution that can serve predictions through a provider-managed endpoint to reduce infrastructure operations. Which option is the strongest fit? The review committee wants a direct mapping from the requirement to the chosen capability. The first release supports 5 departments and is reviewed every 603 days.

1. Asynchronous inference
2. Batch inference job
3. Serverless inference
4. Managed model API
5. Batch inference

Correct answer: D

Why: A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider. It directly addresses the requirement in this scenario.

Option review:

A: Asynchronous inference decouples request submission from result retrieval for longer-running workloads. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Batch execution can be cost-effective when immediate online responses are unnecessary. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Serverless inference automatically provisions serving capacity and is well suited to variable or infrequent traffic when supported. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider. It directly addresses the requirement in this scenario.

E: Batch inference processes many records together and is appropriate when results can be produced on a schedule. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Managed model API – A managed API offloads much of the serving infrastructure, scaling, and maintenance to the service provider.

Question 30

Adventure Works Manufacturing is reviewing a forecasting initiative. The security architect has one primary requirement: give business users an AI-powered workspace that connects to enterprise data and applications and can automate workflows. Which choice best fits the requirement? The solution will serve multiple internal teams, so the recommendation should be reusable without changing the core requirement. The service has a 640-millisecond internal response target for the affected workflow.

1. Amazon Bedrock
2. Kiro
3. Strands Agents
4. Amazon Bedrock AgentCore
5. Amazon Quick

Correct answer: E

Why: Amazon Quick is an AI assistant/workspace for business data, analytics, applications, and agentic tasks. It directly addresses the requirement in this scenario.

Option review:

A: Amazon Bedrock provides managed foundation-model access and GenAI building blocks without requiring customers to manage model-serving infrastructure. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Kiro is an agentic development environment that helps turn prompts and specifications into code, tests, and documentation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Strands Agents is an AWS-backed open-source SDK for building agents. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: AgentCore provides managed infrastructure and services for operating agents at scale. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Amazon Quick is an AI assistant/workspace for business data, analytics, applications, and agentic tasks. It directly addresses the requirement in this scenario.

Learning point: Amazon Quick – Amazon Quick is an AI assistant/workspace for business data, analytics, applications, and agentic tasks.