Topic 01 Practice Test 1 covers GenAI Architecture, Business Requirements, FM Selection, Provider Abstraction, and Resilience in Domain 1 (Tasks 1.1-1.2). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
A claims-assistance team needs a GenAI solution that keeps regulated documents inside approved AWS services, returns answers in under two seconds for interactive users, and can cite internal policy content. Before choosing a model, what should the architect do first? Choose ONE.
- Start with the largest available foundation model
- Define explicit architecture requirements and constraints
- Fine-tune a model immediately
- Create a prompt library first
Correct Answer: B
Correct Answer
Answer B is correct because Define explicit architecture requirements and constraints fits requirements-driven architecture design. It makes latency, data-boundary, grounding, and integration needs measurable before components are selected. Here, requirements-driven architecture design is the governing criterion. Define explicit architecture requirements and constraints therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Start with the largest available foundation model primarily concerns maximizing raw model capability before requirements are known. This scenario needs requirements-driven architecture design. Define explicit architecture requirements and constraints meets that requirement. Start with the largest available foundation model does not, because its focus remains maximizing raw model capability before requirements are known.
Answer C is incorrect because Fine-tune a model immediately primarily concerns adapting model behavior after a justified customization need exists. This scenario needs requirements-driven architecture design. Define explicit architecture requirements and constraints meets that requirement. Fine-tune a model immediately does not, because its focus remains adapting model behavior after a justified customization need exists.
Answer D is incorrect because Create a prompt library first primarily concerns standardizing prompts after solution boundaries and model behavior are understood. This scenario needs requirements-driven architecture design. Define explicit architecture requirements and constraints meets that requirement. Create a prompt library first does not, because its focus remains standardizing prompts after solution boundaries and model behavior are understood.
Question 2
A company is unsure whether an FM can classify support tickets accurately enough before funding a full production build. What should the team implement first? Choose ONE.
- A multi-Region production rollout
- A model retirement workflow
- Run a representative proof of concept
- A provider-abstraction layer only
Correct Answer: C
Correct Answer
Answer C is correct because Run a representative proof of concept fits representative proof-of-concept validation. The uncertainty is feasibility and business value, so a bounded experiment should test the real workload before scale-out. Here, representative proof-of-concept validation is the governing criterion. Run a representative proof of concept therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because A multi-Region production rollout primarily concerns committing full deployment complexity before feasibility is proven. This scenario needs representative proof-of-concept validation. Run a representative proof of concept meets that requirement. A multi-Region production rollout does not, because its focus remains committing full deployment complexity before feasibility is proven.
Answer B is incorrect because A model retirement workflow primarily concerns managing lifecycle after a model has already been adopted. This scenario needs representative proof-of-concept validation. Run a representative proof of concept meets that requirement. A model retirement workflow does not, because its focus remains managing lifecycle after a model has already been adopted.
Answer D is incorrect because A provider-abstraction layer only primarily concerns improving portability without testing whether the use case works. This scenario needs representative proof-of-concept validation. Run a representative proof of concept meets that requirement. A provider-abstraction layer only does not, because its focus remains improving portability without testing whether the use case works.
Question 3
Several teams deploy GenAI workloads with inconsistent security reviews, observability, and failure handling. Which action best supports standardized architecture? Choose ONE.
- Let each team invent its own controls
- Choose one FM for every workload
- Centralize all prompts in one file
- Publish reusable Well-Architected GenAI reference patterns
Correct Answer: D
Correct Answer
Answer D is correct because Publish reusable Well-Architected GenAI reference patterns fits standardized architecture guidance. Shared architecture guidance makes recurring design expectations explicit across teams. Here, standardized architecture guidance is the governing criterion. Publish reusable Well-Architected GenAI reference patterns therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Let each team invent its own controls primarily concerns maximizing local freedom while preserving inconsistency. This scenario needs standardized architecture guidance. Publish reusable Well-Architected GenAI reference patterns meets that requirement. Let each team invent its own controls does not, because its focus remains maximizing local freedom while preserving inconsistency.
Answer B is incorrect because Choose one FM for every workload primarily concerns standardizing a product choice instead of architecture quality. This scenario needs standardized architecture guidance. Publish reusable Well-Architected GenAI reference patterns meets that requirement. Choose one FM for every workload does not, because its focus remains standardizing a product choice instead of architecture quality.
Answer C is incorrect because Centralize all prompts in one file primarily concerns consolidating text assets without addressing architectural controls. This scenario needs standardized architecture guidance. Publish reusable Well-Architected GenAI reference patterns meets that requirement. Centralize all prompts in one file does not, because its focus remains consolidating text assets without addressing architectural controls.
Question 4
A document assistant must accept images and text in the same request. Two candidate FMs have similar language quality, but only one supports the required multimodal input type. What should drive the selection? Choose ONE.
- Choose the model with the lowest text-only benchmark score
- Choose whichever model was evaluated first
- Choose the multimodal-capable model
- Choose the model with the longest provider name
Correct Answer: C
Correct Answer
Answer C is correct because Choose the multimodal-capable model fits capability-fit model selection. Input modality is a hard functional requirement and must be supported before secondary quality differences matter. Here, capability-fit model selection is the governing criterion. Choose the multimodal-capable model therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Choose the model with the lowest text-only benchmark score primarily concerns optimizing a metric that does not establish multimodal support. This scenario needs capability-fit model selection. Choose the multimodal-capable model meets that requirement. Choose the model with the lowest text-only benchmark score does not, because its focus remains optimizing a metric that does not establish multimodal support.
Answer B is incorrect because Choose whichever model was evaluated first primarily concerns using evaluation order rather than capability fit. This scenario needs capability-fit model selection. Choose the multimodal-capable model meets that requirement. Choose whichever model was evaluated first does not, because its focus remains using evaluation order rather than capability fit.
Answer D is incorrect because Choose the model with the longest provider name primarily concerns using irrelevant metadata as a selection criterion. This scenario needs capability-fit model selection. Choose the multimodal-capable model meets that requirement. Choose the model with the longest provider name does not, because its focus remains using irrelevant metadata as a selection criterion.
Question 5
An application team wants to switch between approved FMs without changing business-logic code. Which design best supports that goal? Choose ONE.
- Use a stable provider abstraction interface
- Call provider SDKs directly from every business module
- Store provider names only in comments
- Fine-tune every provider model to the same style
Correct Answer: A
Correct Answer
Answer A is correct because Use a stable provider abstraction interface fits provider-abstraction interface. Business code can call one contract while adapters translate requests for each model provider. Here, provider-abstraction interface is the governing criterion. Use a stable provider abstraction interface therefore satisfies the stated requirement directly.
Incorrect Answers
Answer B is incorrect because Call provider SDKs directly from every business module primarily concerns spreading provider-specific dependencies across the codebase. This scenario needs provider-abstraction interface. Use a stable provider abstraction interface meets that requirement. Call provider SDKs directly from every business module does not, because its focus remains spreading provider-specific dependencies across the codebase.
Answer C is incorrect because Store provider names only in comments primarily concerns documenting choices without changing coupling. This scenario needs provider-abstraction interface. Use a stable provider abstraction interface meets that requirement. Store provider names only in comments does not, because its focus remains documenting choices without changing coupling.
Answer D is incorrect because Fine-tune every provider model to the same style primarily concerns aligning outputs without creating a stable invocation boundary. This scenario needs provider-abstraction interface. Use a stable provider abstraction interface meets that requirement. Fine-tune every provider model to the same style does not, because its focus remains aligning outputs without creating a stable invocation boundary.
Question 6
A production assistant must continue serving requests if the preferred FM becomes unavailable in one Region. The same model is supported through Amazon Bedrock Cross-Region Inference. Which approach best addresses the outage risk? Choose ONE.
- Use Bedrock Cross-Region Inference
- Increase the model temperature
- Store prompts in a second S3 bucket only
- Retry forever against the failed regional endpoint
Correct Answer: A
Correct Answer
Answer A is correct because Use Bedrock Cross-Region Inference fits cross-region inference resilience. The capability can route inference across regions and reduces dependency on a single regional endpoint. Here, cross-region inference resilience is the governing criterion. Use Bedrock Cross-Region Inference therefore satisfies the stated requirement directly.
Incorrect Answers
Answer B is incorrect because Increase the model temperature primarily concerns changing output randomness without improving service availability. This scenario needs cross-region inference resilience. Use Bedrock Cross-Region Inference meets that requirement. Increase the model temperature does not, because its focus remains changing output randomness without improving service availability.
Answer C is incorrect because Store prompts in a second S3 bucket only primarily concerns replicating prompt assets without providing alternate inference capacity. This scenario needs cross-region inference resilience. Use Bedrock Cross-Region Inference meets that requirement. Store prompts in a second S3 bucket only does not, because its focus remains replicating prompt assets without providing alternate inference capacity.
Answer D is incorrect because Retry forever against the failed regional endpoint primarily concerns adding delay while preserving the same failed dependency. This scenario needs cross-region inference resilience. Use Bedrock Cross-Region Inference meets that requirement. Retry forever against the failed regional endpoint does not, because its focus remains adding delay while preserving the same failed dependency.
Question 7
A company has a domain-specific fine-tuned model artifact and needs a managed production endpoint with autoscaling and deployment controls. Which service direction matches the exam objective? Choose ONE.
- Store the model only in Amazon S3
- Deploy the customized model with SageMaker AI
- Use Amazon Route 53 alone
- Put the model name in AWS AppConfig only
Correct Answer: B
Correct Answer
Answer B is correct because Deploy the customized model with SageMaker AI fits managed deployment of a customized model. Sagemaker ai provides managed model hosting patterns appropriate for domain-specific customized models. Here, managed deployment of a customized model is the governing criterion. Deploy the customized model with SageMaker AI therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Store the model only in Amazon S3 primarily concerns persisting the artifact without creating an inference endpoint. This scenario needs managed deployment of a customized model. Deploy the customized model with SageMaker AI meets that requirement. Store the model only in Amazon S3 does not, because its focus remains persisting the artifact without creating an inference endpoint.
Answer C is incorrect because Use Amazon Route 53 alone primarily concerns managing DNS without hosting model inference. This scenario needs managed deployment of a customized model. Deploy the customized model with SageMaker AI meets that requirement. Use Amazon Route 53 alone does not, because its focus remains managing DNS without hosting model inference.
Answer D is incorrect because Put the model name in AWS AppConfig only primarily concerns externalizing configuration without deploying the model. This scenario needs managed deployment of a customized model. Deploy the customized model with SageMaker AI meets that requirement. Put the model name in AWS AppConfig only does not, because its focus remains externalizing configuration without deploying the model.
Question 8
A product team wants a customer-support assistant. The business requires low operational complexity, managed inference, integration with an existing API tier, and no custom model hosting. Which high-level design best matches those constraints? Choose ONE.
- Self-host every model on GPU instances
- Use managed FM inference behind the application layer
- Build a custom training platform before prototyping
- Move all business logic into prompts
Correct Answer: B
Correct Answer
Answer B is correct because Use managed FM inference behind the application layer fits a managed integration architecture. Managed inference satisfies the no-hosting constraint while preserving the application api boundary. Here, a managed integration architecture is the governing criterion. Use managed FM inference behind the application layer therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Self-host every model on GPU instances primarily concerns direct control of model servers and GPU capacity. This scenario needs a managed integration architecture. Use managed FM inference behind the application layer meets that requirement. Self-host every model on GPU instances does not, because its focus remains direct control of model servers and GPU capacity.
Answer C is incorrect because Build a custom training platform before prototyping primarily concerns creating a model-development environment rather than consuming an FM. This scenario needs a managed integration architecture. Use managed FM inference behind the application layer meets that requirement. Build a custom training platform before prototyping does not, because its focus remains creating a model-development environment rather than consuming an FM.
Answer D is incorrect because Move all business logic into prompts primarily concerns reducing application code by relying on prompt instructions. This scenario needs a managed integration architecture. Use managed FM inference behind the application layer meets that requirement. Move all business logic into prompts does not, because its focus remains reducing application code by relying on prompt instructions.
Question 9
A prototype summarizes call transcripts well, but leadership wants evidence that it will meet the production latency target. Which next proof-of-concept activity is most useful? Choose ONE.
- Add more UI themes
- Fine-tune for company terminology first
- Measure end-to-end latency under realistic load
- Expand to every business unit immediately
Correct Answer: C
Correct Answer
Answer C is correct because Measure end-to-end latency under realistic load fits performance validation in the proof of concept. Production feasibility depends on observed response time under representative inputs and request volume. Here, performance validation in the proof of concept is the governing criterion. Measure end-to-end latency under realistic load therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Add more UI themes primarily concerns changing presentation without testing the latency risk. This scenario needs performance validation in the proof of concept. Measure end-to-end latency under realistic load meets that requirement. Add more UI themes does not, because its focus remains changing presentation without testing the latency risk.
Answer B is incorrect because Fine-tune for company terminology first primarily concerns changing model behavior before isolating the stated performance concern. This scenario needs performance validation in the proof of concept. Measure end-to-end latency under realistic load meets that requirement. Fine-tune for company terminology first does not, because its focus remains changing model behavior before isolating the stated performance concern.
Answer D is incorrect because Expand to every business unit immediately primarily concerns increasing scope before the performance assumption is validated. This scenario needs performance validation in the proof of concept. Measure end-to-end latency under realistic load meets that requirement. Expand to every business unit immediately does not, because its focus remains increasing scope before the performance assumption is validated.
Question 10
A coding assistant needs reliable structured JSON output for downstream automation. Which model-selection activity is most appropriate? Choose ONE.
- Compare only public general-knowledge scores
- Select by maximum context window alone
- Choose a model based only on launch date
- Benchmark representative structured-output tasks and failures
Correct Answer: D
Correct Answer
Answer D is correct because Benchmark representative structured-output tasks and failures fits task-specific model benchmarking. The team needs evidence that the model follows the required response contract on realistic inputs. Here, task-specific model benchmarking is the governing criterion. Benchmark representative structured-output tasks and failures therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Compare only public general-knowledge scores primarily concerns using broad benchmarks that do not test the required output behavior. This scenario needs task-specific model benchmarking. Benchmark representative structured-output tasks and failures meets that requirement. Compare only public general-knowledge scores does not, because its focus remains using broad benchmarks that do not test the required output behavior.
Answer B is incorrect because Select by maximum context window alone primarily concerns optimizing input capacity rather than JSON reliability. This scenario needs task-specific model benchmarking. Benchmark representative structured-output tasks and failures meets that requirement. Select by maximum context window alone does not, because its focus remains optimizing input capacity rather than JSON reliability.
Answer C is incorrect because Choose a model based only on launch date primarily concerns treating recency as a substitute for workload evidence. This scenario needs task-specific model benchmarking. Benchmark representative structured-output tasks and failures meets that requirement. Choose a model based only on launch date does not, because its focus remains treating recency as a substitute for workload evidence.
Question 11
A platform must change the preferred FM at runtime without redeploying the application. Which configuration pattern is most appropriate? Choose ONE.
- Hard-code the model ID in source code
- Encode the model ID in the user prompt
- Externalize model routing in AWS AppConfig
- Rename the Lambda function when models change
Correct Answer: C
Correct Answer
Answer C is correct because Externalize model routing in AWS AppConfig fits configuration-driven model switching. Externalized configuration allows controlled runtime selection without editing application code. Here, configuration-driven model switching is the governing criterion. Externalize model routing in AWS AppConfig therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Hard-code the model ID in source code primarily concerns requiring code modification and deployment for every switch. This scenario needs configuration-driven model switching. Externalize model routing in AWS AppConfig meets that requirement. Hard-code the model ID in source code does not, because its focus remains requiring code modification and deployment for every switch.
Answer B is incorrect because Encode the model ID in the user prompt primarily concerns mixing infrastructure routing with user content. This scenario needs configuration-driven model switching. Externalize model routing in AWS AppConfig meets that requirement. Encode the model ID in the user prompt does not, because its focus remains mixing infrastructure routing with user content.
Answer D is incorrect because Rename the Lambda function when models change primarily concerns changing resource labels rather than routing configuration. This scenario needs configuration-driven model switching. Externalize model routing in AWS AppConfig meets that requirement. Rename the Lambda function when models change does not, because its focus remains changing resource labels rather than routing configuration.
Question 12
An application sees repeated FM timeouts. The team wants to stop hammering the failing dependency, wait, and then probe recovery. Which pattern should it implement? Choose ONE.
- A larger context window
- A prompt template version
- Use a circuit breaker around FM calls
- An embedding index rebuild
Correct Answer: C
Correct Answer
Answer C is correct because Use a circuit breaker around FM calls fits circuit-breaker failure containment. The pattern opens after failures, limits repeated calls, and allows controlled recovery checks. Here, circuit-breaker failure containment is the governing criterion. Use a circuit breaker around FM calls therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because A larger context window primarily concerns changing input capacity rather than handling dependency failures. This scenario needs circuit-breaker failure containment. Use a circuit breaker around FM calls meets that requirement. A larger context window does not, because its focus remains changing input capacity rather than handling dependency failures.
Answer B is incorrect because A prompt template version primarily concerns managing prompt text rather than invocation health. This scenario needs circuit-breaker failure containment. Use a circuit breaker around FM calls meets that requirement. A prompt template version does not, because its focus remains managing prompt text rather than invocation health.
Answer D is incorrect because An embedding index rebuild primarily concerns refreshing retrieval data instead of controlling failing calls. This scenario needs circuit-breaker failure containment. Use a circuit breaker around FM calls meets that requirement. An embedding index rebuild does not, because its focus remains refreshing retrieval data instead of controlling failing calls.
Question 13
A team needs to adapt a large model to domain terminology while changing a small subset of trainable parameters rather than updating every model weight. Which technique best matches the requirement? Choose ONE.
- Use parameter-efficient adaptation such as LoRA
- Full model retraining from random initialization
- Cross-Region Inference
- API Gateway request mapping
Correct Answer: A
Correct Answer
Answer A is correct because Use parameter-efficient adaptation such as LoRA fits parameter-efficient model adaptation. Lora is designed to adapt model behavior with a much smaller trainable parameter set than full fine-tuning. Here, parameter-efficient model adaptation is the governing criterion. Use parameter-efficient adaptation such as LoRA therefore satisfies the stated requirement directly.
Incorrect Answers
Answer B is incorrect because Full model retraining from random initialization primarily concerns rebuilding all model weights rather than adapting efficiently. This scenario needs parameter-efficient model adaptation. Use parameter-efficient adaptation such as LoRA meets that requirement. Full model retraining from random initialization does not, because its focus remains rebuilding all model weights rather than adapting efficiently.
Answer C is incorrect because Cross-Region Inference primarily concerns improving regional inference resilience rather than customizing model parameters. This scenario needs parameter-efficient model adaptation. Use parameter-efficient adaptation such as LoRA meets that requirement. Cross-Region Inference does not, because its focus remains improving regional inference resilience rather than customizing model parameters.
Answer D is incorrect because API Gateway request mapping primarily concerns transforming API payloads rather than adapting a model. This scenario needs parameter-efficient model adaptation. Use parameter-efficient adaptation such as LoRA meets that requirement. API Gateway request mapping does not, because its focus remains transforming API payloads rather than adapting a model.
Question 14
A legal-document assistant must support synchronous chat for short questions and asynchronous processing for thousand-page case files. Which architectural decision best addresses the workload difference? Choose ONE.
- Separate real-time chat from asynchronous document jobs
- Force both workloads through one synchronous endpoint
- Use only scheduled batch processing
- Fine-tune the FM for longer documents
Correct Answer: A
Correct Answer
Answer A is correct because Separate real-time chat from asynchronous document jobs fits workload-specific integration patterns. The workloads have different latency and execution-duration characteristics that should not share one request path. Here, workload-specific integration patterns is the governing criterion. Separate real-time chat from asynchronous document jobs therefore satisfies the stated requirement directly.
Incorrect Answers
Answer B is incorrect because Force both workloads through one synchronous endpoint primarily concerns using one request-response pattern for operational simplicity. This scenario needs workload-specific integration patterns. Separate real-time chat from asynchronous document jobs meets that requirement. Force both workloads through one synchronous endpoint does not, because its focus remains using one request-response pattern for operational simplicity.
Answer C is incorrect because Use only scheduled batch processing primarily concerns processing all work offline regardless of user latency needs. This scenario needs workload-specific integration patterns. Separate real-time chat from asynchronous document jobs meets that requirement. Use only scheduled batch processing does not, because its focus remains processing all work offline regardless of user latency needs.
Answer D is incorrect because Fine-tune the FM for longer documents primarily concerns adapting model behavior rather than changing execution patterns. This scenario needs workload-specific integration patterns. Separate real-time chat from asynchronous document jobs meets that requirement. Fine-tune the FM for longer documents does not, because its focus remains adapting model behavior rather than changing execution patterns.
Question 15
A customer assistant has a strict p95 latency target. Two models meet quality requirements, but one is consistently slower under the expected request size. Which selection principle applies? Choose ONE.
- Prefer the slower model because it is larger
- Increase the user timeout without approval
- Ignore latency until production
- Use measured representative-workload latency
Correct Answer: D
Correct Answer
Answer D is correct because Use measured representative-workload latency fits latency-aware model selection. Once quality is acceptable, the explicit service-level requirement should determine which model is operationally viable. Here, latency-aware model selection is the governing criterion. Use measured representative-workload latency therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Prefer the slower model because it is larger primarily concerns assuming size overrides an explicit latency requirement. This scenario needs latency-aware model selection. Use measured representative-workload latency meets that requirement. Prefer the slower model because it is larger does not, because its focus remains assuming size overrides an explicit latency requirement.
Answer B is incorrect because Increase the user timeout without approval primarily concerns changing the requirement instead of evaluating against it. This scenario needs latency-aware model selection. Use measured representative-workload latency meets that requirement. Increase the user timeout without approval does not, because its focus remains changing the requirement instead of evaluating against it.
Answer C is incorrect because Ignore latency until production primarily concerns deferring a known acceptance criterion until after selection. This scenario needs latency-aware model selection. Use measured representative-workload latency meets that requirement. Ignore latency until production does not, because its focus remains deferring a known acceptance criterion until after selection.
Question 16
Different FMs require different request and response schemas, but the product API should remain stable. What component should absorb those differences? Choose ONE.
- Normalize provider payloads with adapters
- Every UI client should understand every model schema
- The user should select a provider-specific JSON format
- A larger FM should normalize the API contract
Correct Answer: A
Correct Answer
Answer A is correct because Normalize provider payloads with adapters fits request-response normalization adapters. Translation at the provider boundary prevents model schema differences from leaking into product logic. Here, request-response normalization adapters is the governing criterion. Normalize provider payloads with adapters therefore satisfies the stated requirement directly.
Incorrect Answers
Answer B is incorrect because Every UI client should understand every model schema primarily concerns moving provider coupling into all consumers. This scenario needs request-response normalization adapters. Normalize provider payloads with adapters meets that requirement. Every UI client should understand every model schema does not, because its focus remains moving provider coupling into all consumers.
Answer C is incorrect because The user should select a provider-specific JSON format primarily concerns exposing infrastructure variation as a user responsibility. This scenario needs request-response normalization adapters. Normalize provider payloads with adapters meets that requirement. The user should select a provider-specific JSON format does not, because its focus remains exposing infrastructure variation as a user responsibility.
Answer D is incorrect because A larger FM should normalize the API contract primarily concerns using generation rather than deterministic integration code. This scenario needs request-response normalization adapters. Normalize provider payloads with adapters meets that requirement. A larger FM should normalize the API contract does not, because its focus remains using generation rather than deterministic integration code.
Question 17
The primary FM is unavailable. The business permits shorter, lower-quality responses from a secondary model instead of an outage. Which resilience strategy best fits? Choose ONE.
- Return fabricated answers from cached UI text
- Disable all requests until the primary returns
- Fine-tune the failed model during the incident
- Fail over to an approved reduced-capability model
Correct Answer: D
Correct Answer
Answer D is correct because Fail over to an approved reduced-capability model fits graceful model fallback. The service preserves core availability while making an explicit quality trade-off allowed by the business. Here, graceful model fallback is the governing criterion. Fail over to an approved reduced-capability model therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Return fabricated answers from cached UI text primarily concerns maintaining appearance without reliable model output. This scenario needs graceful model fallback. Fail over to an approved reduced-capability model meets that requirement. Return fabricated answers from cached UI text does not, because its focus remains maintaining appearance without reliable model output.
Answer B is incorrect because Disable all requests until the primary returns primarily concerns ignoring the approved degraded-service option. This scenario needs graceful model fallback. Fail over to an approved reduced-capability model meets that requirement. Disable all requests until the primary returns does not, because its focus remains ignoring the approved degraded-service option.
Answer C is incorrect because Fine-tune the failed model during the incident primarily concerns performing customization that does not restore endpoint availability. This scenario needs graceful model fallback. Fail over to an approved reduced-capability model meets that requirement. Fine-tune the failed model during the incident does not, because its focus remains performing customization that does not restore endpoint availability.
Question 18
A platform maintains several approved customized model versions and must know which artifact is ready for staging versus production. Which capability is most appropriate? Choose ONE.
- Store version names only in prompt text
- Version approved models in SageMaker Model Registry
- Use CloudFront cache behaviors
- Use security-group descriptions
Correct Answer: B
Correct Answer
Answer B is correct because Version approved models in SageMaker Model Registry fits versioned customized-model governance. The registry provides lifecycle metadata and promotion state for managed model versions. Here, versioned customized-model governance is the governing criterion. Version approved models in SageMaker Model Registry therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Store version names only in prompt text primarily concerns mixing deployment metadata with model instructions. This scenario needs versioned customized-model governance. Version approved models in SageMaker Model Registry meets that requirement. Store version names only in prompt text does not, because its focus remains mixing deployment metadata with model instructions.
Answer C is incorrect because Use CloudFront cache behaviors primarily concerns controlling content delivery rather than model version lifecycle. This scenario needs versioned customized-model governance. Version approved models in SageMaker Model Registry meets that requirement. Use CloudFront cache behaviors does not, because its focus remains controlling content delivery rather than model version lifecycle.
Answer D is incorrect because Use security-group descriptions primarily concerns recording network notes rather than deployment approval state. This scenario needs versioned customized-model governance. Version approved models in SageMaker Model Registry meets that requirement. Use security-group descriptions does not, because its focus remains recording network notes rather than deployment approval state.
Question 19
A multinational team is planning a GenAI application. Some countries require processing in approved Regions, while other users prioritize lowest latency. Which design activity is most important before selecting deployment Regions? Choose ONE.
- Choose the cheapest Region globally
- Map regulatory and latency needs to Regions
- Use a single Region for every user
- Increase model temperature for distant users
Correct Answer: B
Correct Answer
Answer B is correct because Map regulatory and latency needs to Regions fits regional constraint mapping. The architecture must satisfy data-location obligations and user experience constraints together. Here, regional constraint mapping is the governing criterion. Map regulatory and latency needs to Regions therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because Choose the cheapest Region globally primarily concerns optimizing price without first validating regulatory eligibility. This scenario needs regional constraint mapping. Map regulatory and latency needs to Regions meets that requirement. Choose the cheapest Region globally does not, because its focus remains optimizing price without first validating regulatory eligibility.
Answer C is incorrect because Use a single Region for every user primarily concerns simplifying operations at the expense of regional requirements. This scenario needs regional constraint mapping. Map regulatory and latency needs to Regions meets that requirement. Use a single Region for every user does not, because its focus remains simplifying operations at the expense of regional requirements.
Answer D is incorrect because Increase model temperature for distant users primarily concerns changing generation variability rather than deployment location. This scenario needs regional constraint mapping. Map regulatory and latency needs to Regions meets that requirement. Increase model temperature for distant users does not, because its focus remains changing generation variability rather than deployment location.
Question 20
A new customized model version passes evaluation and should move automatically from test to staging using repeatable controls. What should the team implement? Choose ONE.
- A manual copy procedure in a wiki
- A higher inference temperature
- A new prompt for every deployment
- Automate deployment with explicit promotion gates
Correct Answer: D
Correct Answer
Answer D is correct because Automate deployment with explicit promotion gates fits automated model promotion. Repeatable automation reduces manual drift while preserving approval checks between lifecycle stages. Here, automated model promotion is the governing criterion. Automate deployment with explicit promotion gates therefore satisfies the stated requirement directly.
Incorrect Answers
Answer A is incorrect because A manual copy procedure in a wiki primarily concerns depending on human steps for each deployment. This scenario needs automated model promotion. Automate deployment with explicit promotion gates meets that requirement. A manual copy procedure in a wiki does not, because its focus remains depending on human steps for each deployment.
Answer B is incorrect because A higher inference temperature primarily concerns changing generation behavior rather than release automation. This scenario needs automated model promotion. Automate deployment with explicit promotion gates meets that requirement. A higher inference temperature does not, because its focus remains changing generation behavior rather than release automation.
Answer C is incorrect because A new prompt for every deployment primarily concerns coupling model release to unrelated prompt changes. This scenario needs automated model promotion. Automate deployment with explicit promotion gates meets that requirement. A new prompt for every deployment does not, because its focus remains coupling model release to unrelated prompt changes.
