Amazon AWS AI Practitioner AIF-C01 Foundation Model Selection Inference Parameters And RAG Practice Test

 

AIF-C01 skills 3.1 | 30 original questions

This AWS Certified AI Practitioner AIF-C01 practice test focuses on foundation model selection inference parameters and rag 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 security architect at Tailspin Toys is preparing a recommendation for a customer-support modernization. The recommendation must cap response length to control latency and output-token cost. 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 7 candidate designs after a 245-day proof of concept.

1. Use prompt caching when supported and repeated context is large
2. Limit output length when long responses are not needed
3. Lower temperature
4. Higher temperature
5. Shorter maximum output

Correct answer: E

Why: Limiting generated length can reduce both wait time and token spend. It directly addresses the requirement in this scenario.

Option review:

A: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Higher temperature generally increases randomness and variation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Limiting generated length can reduce both wait time and token spend. It directly addresses the requirement in this scenario.

Learning point: Shorter maximum output – Limiting generated length can reduce both wait time and token spend.

Question 2

City Power and Light has completed discovery for a agentic workflow trial. Before implementation, the risk manager must decide how to return supporting source passages with an answer so users can verify where the information came from. Which choice best satisfies that requirement? The team will validate the result with representative production examples before rollout. The control owner requires evidence from 4 test groups before the 282-day release review.

1. RAG citation and grounding
2. Confidence scoring
3. Retrieval Augmented Generation (RAG)
4. RAG grounding
5. Human verification for high-impact claims

Correct answer: A

Why: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. It directly addresses the requirement in this scenario.

Option review:

A: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. It directly addresses the requirement in this scenario.

B: Confidence signals can support routing and human review, but they must be validated for the actual application. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: RAG retrieves relevant content at request time and adds it to the model context. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Grounding with RAG can reduce unsupported generation when retrieval sources are relevant and trustworthy. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Human review is appropriate when residual hallucination risk is unacceptable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: RAG citation and grounding – Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable.

Question 3

While planning a contact-center transformation, Consolidated Messenger identifies this requirement: select a model that can natively accept the required combination of text and images. Which option should the security architect prioritize if the goal is to keep the design easy to explain? The pilot has representative data, and the team will measure the selected approach against an agreed acceptance threshold. The project has 9 downstream consumers and a monthly review of approximately 319 sampled interactions.

1. Customization support
2. Multimodal model
3. Cost
4. Customizable model
5. Compliance fit

Correct answer: B

Why: Modality support must match application inputs and outputs. It directly addresses the requirement in this scenario.

Option review:

A: Required customization capabilities narrow the set of suitable models. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Modality support must match application inputs and outputs. It directly addresses the requirement in this scenario.

C: Model economics should be evaluated at expected production scale. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Customization requirements can determine model eligibility. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Compliance constraints can be hard gates in model selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Multimodal model – Modality support must match application inputs and outputs.

Question 4

A proof of concept at Nod Publishers exposed a design decision for the risk manager: the solution must remove irrelevant context to reduce token cost and avoid distracting the model. 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 6 application teams, each using the same approved requirement set for the next 356 days.

1. Limit output length when long responses are not needed
2. Account for both input and output token rates
3. Shorter input context
4. Shorter maximum output
5. Lower temperature

Correct answer: C

Why: Reducing unnecessary input can improve efficiency and sometimes quality. It directly addresses the requirement in this scenario.

Option review:

A: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Reducing unnecessary input can improve efficiency and sometimes quality. It directly addresses the requirement in this scenario.

D: Limiting generated length can reduce both wait time and token spend. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Shorter input context – Reducing unnecessary input can improve efficiency and sometimes quality.

Question 5

Fabrikam Health is documenting the target state for a sales-assistant rollout. The security architect needs a solution that can make frequently changing policy documents available to answers without retraining model weights. 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 3 business workflows and 393 recent production cases.

1. RAG citation and grounding
2. Amazon Bedrock Knowledge Bases
3. RAG grounding
4. Update the knowledge source instead of fine-tuning
5. Confidence scoring

Correct answer: D

Why: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. It directly addresses the requirement in this scenario.

Option review:

A: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Grounding with RAG can reduce unsupported generation when retrieval sources are relevant and trustworthy. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. It directly addresses the requirement in this scenario.

E: Confidence signals can support routing and human review, but they must be validated for the actual application. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Update the knowledge source instead of fine-tuning – RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model.

Question 6

Wingtip Logistics is reviewing a internal search upgrade. The risk manager has one primary requirement: use a compact model when cost, speed, or deployment constraints matter and quality remains acceptable. 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 430 internal users across 8 business units.

1. Low-latency model
2. Compliance fit
3. Multilingual model
4. Latency
5. Smaller model

Correct answer: E

Why: Smaller models can improve economics and latency. It directly addresses the requirement in this scenario.

Option review:

A: Latency requirements can outweigh marginal benchmark gains from a larger model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Compliance constraints can be hard gates in model selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Language coverage and quality should be tested for the actual user population. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Larger or more complex models can provide quality gains at the cost of latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Smaller models can improve economics and latency. It directly addresses the requirement in this scenario.

Learning point: Smaller model – Smaller models can improve economics and latency.

Question 7

During a design review for Trey Research, the security architect must make text generation more consistent and less diverse when the task rewards predictable wording. The team also wants to control recurring cost. 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 467 requests during its busiest hour and has a documented fallback path.

1. Lower temperature
2. Shorter maximum output
3. Shorter input context
4. Use prompt caching when supported and repeated context is large
5. Higher temperature

Correct answer: A

Why: Lower temperature generally reduces randomness in token selection. It directly addresses the requirement in this scenario.

Option review:

A: Lower temperature generally reduces randomness in token selection. It directly addresses the requirement in this scenario.

B: Limiting generated length can reduce both wait time and token spend. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Reducing unnecessary input can improve efficiency and sometimes quality. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Higher temperature generally increases randomness and variation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Lower temperature – Lower temperature generally reduces randomness in token selection.

Question 8

Bellows College is moving a knowledge-assistant rollout from pilot to production. The key decision is how to ground a model response in retrieved enterprise documents without changing the model weights. Which option is the strongest fit if the team wants to meet a strict latency target? Operational ownership is already assigned, so the team is comparing technical fit rather than staffing models. The pilot uses 504 representative records from 2 approved data sources.

1. Selective context retrieval
2. Retrieval Augmented Generation (RAG)
3. Tool-result injection
4. Output validation
5. Update the knowledge source instead of fine-tuning

Correct answer: B

Why: RAG retrieves relevant content at request time and adds it to the model context. It directly addresses the requirement in this scenario.

Option review:

A: Selective retrieval improves signal-to-noise and controls token usage. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: RAG retrieves relevant content at request time and adds it to the model context. It directly addresses the requirement in this scenario.

C: Agentic applications use tool outputs as context so the model can continue from authoritative external information. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Validation catches outputs that are inconsistent with authoritative data or format requirements. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Retrieval Augmented Generation (RAG) – RAG retrieves relevant content at request time and adds it to the model context.

Question 9

A workshop at Blue Yonder Airlines focuses on a single decision: how to prefer a model/service feature that can reuse a large stable prompt prefix across many calls. Which option should the security architect recommend? Existing application interfaces can accommodate any of the listed choices, so functional fit is the deciding factor. The first release supports 7 departments and is reviewed every 541 days.

1. Modality support
2. Low-latency model
3. Prompt-caching support
4. Customizable model
5. Lower-cost model

Correct answer: C

Why: Prompt caching can improve economics and latency for repeated context. It directly addresses the requirement in this scenario.

Option review:

A: A model must support the required modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Latency requirements can outweigh marginal benchmark gains from a larger model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Prompt caching can improve economics and latency for repeated context. It directly addresses the requirement in this scenario.

D: Customization requirements can determine model eligibility. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Cost is a first-class model-selection criterion once minimum quality and governance gates are met. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Prompt-caching support – Prompt caching can improve economics and latency for repeated context.

Question 10

For the personalization program at Woodgrove Bank, stakeholders need to increase output diversity and creativity when some variability is acceptable. 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 578-millisecond internal response target for the affected workflow.

1. Shorter input context
2. Lower temperature
3. Account for both input and output token rates
4. Higher temperature
5. Reduce unnecessary input tokens

Correct answer: D

Why: Higher temperature generally increases randomness and variation. It directly addresses the requirement in this scenario.

Option review:

A: Reducing unnecessary input can improve efficiency and sometimes quality. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Higher temperature generally increases randomness and variation. It directly addresses the requirement in this scenario.

E: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Higher temperature – Higher temperature generally increases randomness and variation.

Question 11

Wide World Importers is comparing alternatives for its developer-productivity pilot. The security architect needs to implement managed retrieval for Bedrock applications over configured enterprise data sources. Which option is most appropriate while trying to use current managed AWS capabilities? The review committee wants a direct mapping from the requirement to the chosen capability. The team is comparing 9 candidate designs after a 615-day proof of concept.

1. Confidence scoring
2. RAG grounding
3. RAG citation and grounding
4. Conversation-state management
5. Amazon Bedrock Knowledge Bases

Correct answer: E

Why: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. It directly addresses the requirement in this scenario.

Option review:

A: Confidence signals can support routing and human review, but they must be validated for the actual application. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Grounding with RAG can reduce unsupported generation when retrieval sources are relevant and trustworthy. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Managing conversation state balances continuity with context-window and cost constraints. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. It directly addresses the requirement in this scenario.

Learning point: Amazon Bedrock Knowledge Bases – Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG.

Question 12

An architecture review at VanArsdel Media has narrowed a fraud-review pilot decision to one requirement: choose a model optimized for faster responses when interactive user experience has a strict latency budget. What should the risk manager 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 6 test groups before the 652-day release review.

1. Low-latency model
2. Modality support
3. Long-context model
4. Cost
5. Customization support

Correct answer: A

Why: Latency requirements can outweigh marginal benchmark gains from a larger model. It directly addresses the requirement in this scenario.

Option review:

A: Latency requirements can outweigh marginal benchmark gains from a larger model. It directly addresses the requirement in this scenario.

B: A model must support the required modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Context limits constrain how much material can be handled in one interaction. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Model economics should be evaluated at expected production scale. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Required customization capabilities narrow the set of suitable models. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Low-latency model – Latency requirements can outweigh marginal benchmark gains from a larger model.

Question 13

The security architect at Datum Dynamics is preparing a recommendation for a analytics modernization. The recommendation must cap response length to control latency and output-token cost. 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 3 downstream consumers and a monthly review of approximately 689 sampled interactions.

1. Account for both input and output token rates
2. Shorter maximum output
3. Limit output length when long responses are not needed
4. Shorter input context
5. Lower temperature

Correct answer: B

Why: Limiting generated length can reduce both wait time and token spend. It directly addresses the requirement in this scenario.

Option review:

A: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Limiting generated length can reduce both wait time and token spend. It directly addresses the requirement in this scenario.

C: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Reducing unnecessary input can improve efficiency and sometimes quality. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Shorter maximum output – Limiting generated length can reduce both wait time and token spend.

Question 14

Alpine Ski House has completed discovery for a compliance-assistant prototype. Before implementation, the risk manager must decide how to return supporting source passages with an answer so users can verify where the information came from. 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 8 application teams, each using the same approved requirement set for the next 726 days.

1. Update the knowledge source instead of fine-tuning
2. Human verification for high-impact claims
3. RAG citation and grounding
4. Retrieval Augmented Generation (RAG)
5. Conversation-state management

Correct answer: C

Why: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. It directly addresses the requirement in this scenario.

Option review:

A: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Human review is appropriate when residual hallucination risk is unacceptable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. It directly addresses the requirement in this scenario.

D: RAG retrieves relevant content at request time and adds it to the model context. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Managing conversation state balances continuity with context-window and cost constraints. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: RAG citation and grounding – Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable.

Question 15

While planning a forecasting initiative, Humongous Insurance identifies this requirement: avoid unnecessary model complexity when a simpler candidate passes application-specific evaluation. Which option should the security architect prioritize if the goal is to keep the design easy to explain? The recommendation must solve the stated requirement without introducing unrelated platform complexity. The evaluation set contains examples from 5 business workflows and 763 recent production cases.

1. Modality support
2. Multilingual model
3. Smaller model
4. Simpler model architecture for the task
5. Low-latency model

Correct answer: D

Why: Model complexity should be justified by measurable value. It directly addresses the requirement in this scenario.

Option review:

A: A model must support the required modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Language coverage and quality should be tested for the actual user population. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Smaller models can improve economics and latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Model complexity should be justified by measurable value. It directly addresses the requirement in this scenario.

E: Latency requirements can outweigh marginal benchmark gains from a larger model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Simpler model architecture for the task – Model complexity should be justified by measurable value.

Question 16

A proof of concept at Graphic Design Institute exposed a design decision for the risk manager: the solution must make text generation more consistent and less diverse when the task rewards predictable wording. 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 800 internal users across 2 business units.

1. Account for both input and output token rates
2. Use prompt caching when supported and repeated context is large
3. Limit output length when long responses are not needed
4. Higher temperature
5. Lower temperature

Correct answer: E

Why: Lower temperature generally reduces randomness in token selection. It directly addresses the requirement in this scenario.

Option review:

A: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Higher temperature generally increases randomness and variation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Lower temperature generally reduces randomness in token selection. It directly addresses the requirement in this scenario.

Learning point: Lower temperature – Lower temperature generally reduces randomness in token selection.

Question 17

Relecloud is documenting the target state for a agentic workflow trial. The security architect needs a solution that can ground a model response in retrieved enterprise documents without changing the model weights. 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 837 requests during its busiest hour and has a documented fallback path.

1. Retrieval Augmented Generation (RAG)
2. RAG grounding
3. Tool-result injection
4. Amazon Bedrock Knowledge Bases
5. Human verification for high-impact claims

Correct answer: A

Why: RAG retrieves relevant content at request time and adds it to the model context. It directly addresses the requirement in this scenario.

Option review:

A: RAG retrieves relevant content at request time and adds it to the model context. It directly addresses the requirement in this scenario.

B: Grounding with RAG can reduce unsupported generation when retrieval sources are relevant and trustworthy. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Agentic applications use tool outputs as context so the model can continue from authoritative external information. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Human review is appropriate when residual hallucination risk is unacceptable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Retrieval Augmented Generation (RAG) – RAG retrieves relevant content at request time and adds it to the model context.

Question 18

Adventure Works Manufacturing is reviewing a contact-center transformation. The risk manager has one primary requirement: select a model with sufficient input and output limits for the expected context size. 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 874 representative records from 4 approved data sources.

1. Multilingual model
2. Long-context model
3. Customizable model
4. Quality and capability
5. Context-window constraints

Correct answer: B

Why: Context limits constrain how much material can be handled in one interaction. It directly addresses the requirement in this scenario.

Option review:

A: Language coverage and quality should be tested for the actual user population. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Context limits constrain how much material can be handled in one interaction. It directly addresses the requirement in this scenario.

C: Customization requirements can determine model eligibility. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Benchmarks and application-specific evaluations should guide capability selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Input and output limits can constrain application design. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Long-context model – Context limits constrain how much material can be handled in one interaction.

Question 19

During a design review for Proseware Services, the security architect must increase output diversity and creativity when some variability is acceptable. The team also wants to control recurring cost. 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 9 departments and is reviewed every 911 days.

1. Lower temperature
2. Shorter input context
3. Higher temperature
4. Use prompt caching when supported and repeated context is large
5. Reduce unnecessary input tokens

Correct answer: C

Why: Higher temperature generally increases randomness and variation. It directly addresses the requirement in this scenario.

Option review:

A: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Reducing unnecessary input can improve efficiency and sometimes quality. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Higher temperature generally increases randomness and variation. It directly addresses the requirement in this scenario.

D: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Higher temperature – Higher temperature generally increases randomness and variation.

Question 20

Lucerne Publishing is moving a sales-assistant rollout from pilot to production. The key decision is how to implement managed retrieval for Bedrock applications over configured enterprise data sources. Which option is the strongest fit if the team wants to meet a strict latency target? The team will validate the result with representative production examples before rollout. The service has a 948-millisecond internal response target for the affected workflow.

1. Update the knowledge source instead of fine-tuning
2. Context engineering
3. Confidence scoring
4. Amazon Bedrock Knowledge Bases
5. Conversation-state management

Correct answer: D

Why: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. It directly addresses the requirement in this scenario.

Option review:

A: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Context engineering manages the information supplied to a model so it can solve the current task with the right grounding. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Confidence signals can support routing and human review, but they must be validated for the actual application. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. It directly addresses the requirement in this scenario.

E: Managing conversation state balances continuity with context-window and cost constraints. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Amazon Bedrock Knowledge Bases – Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG.

Question 21

A workshop at Lamna Healthcare focuses on a single decision: how to select a model evaluated for the languages the application must support. Which option should the security architect recommend? The pilot has representative data, and the team will measure the selected approach against an agreed acceptance threshold. The team is comparing 3 candidate designs after a 985-day proof of concept.

1. Lower-cost model
2. Model complexity
3. Latency
4. Customizable model
5. Multilingual model

Correct answer: E

Why: Language coverage and quality should be tested for the actual user population. It directly addresses the requirement in this scenario.

Option review:

A: Cost is a first-class model-selection criterion once minimum quality and governance gates are met. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Complexity can raise cost, latency, and operational burden. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Larger or more complex models can provide quality gains at the cost of latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Customization requirements can determine model eligibility. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Language coverage and quality should be tested for the actual user population. It directly addresses the requirement in this scenario.

Learning point: Multilingual model – Language coverage and quality should be tested for the actual user population.

Question 22

For the document-intelligence project at Contoso Retail, stakeholders need to remove irrelevant context to reduce token cost and avoid distracting the model. 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 8 test groups before the 62-day release review.

1. Shorter input context
2. Limit output length when long responses are not needed
3. Shorter maximum output
4. Account for both input and output token rates
5. Higher temperature

Correct answer: A

Why: Reducing unnecessary input can improve efficiency and sometimes quality. It directly addresses the requirement in this scenario.

Option review:

A: Reducing unnecessary input can improve efficiency and sometimes quality. It directly addresses the requirement in this scenario.

B: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Limiting generated length can reduce both wait time and token spend. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Higher temperature generally increases randomness and variation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Shorter input context – Reducing unnecessary input can improve efficiency and sometimes quality.

Question 23

Fourth Coffee is comparing alternatives for its knowledge-assistant rollout. The security architect needs to make frequently changing policy documents available to answers without retraining model weights. Which option is most appropriate while trying to use current managed AWS capabilities? The team wants the least complex technically correct choice that satisfies the requirement. The project has 5 downstream consumers and a monthly review of approximately 99 sampled interactions.

1. Retrieval Augmented Generation (RAG)
2. Update the knowledge source instead of fine-tuning
3. Human verification for high-impact claims
4. Context engineering
5. Confidence scoring

Correct answer: B

Why: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. It directly addresses the requirement in this scenario.

Option review:

A: RAG retrieves relevant content at request time and adds it to the model context. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. It directly addresses the requirement in this scenario.

C: Human review is appropriate when residual hallucination risk is unacceptable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Context engineering manages the information supplied to a model so it can solve the current task with the right grounding. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Confidence signals can support routing and human review, but they must be validated for the actual application. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Update the knowledge source instead of fine-tuning – RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model.

Question 24

An architecture review at Margie Travel has narrowed a claims-processing redesign decision to one requirement: select a model that supports the required fine-tuning or adaptation method. What should the risk manager select? The workload has passed basic feasibility checks, so the remaining question is which approach best matches the requirement. The rollout spans 2 application teams, each using the same approved requirement set for the next 136 days.

1. Lower-cost model
2. Quality and capability
3. Customizable model
4. Customization support
5. Smaller model

Correct answer: C

Why: Customization requirements can determine model eligibility. It directly addresses the requirement in this scenario.

Option review:

A: Cost is a first-class model-selection criterion once minimum quality and governance gates are met. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Benchmarks and application-specific evaluations should guide capability selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Customization requirements can determine model eligibility. It directly addresses the requirement in this scenario.

D: Required customization capabilities narrow the set of suitable models. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Smaller models can improve economics and latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Customizable model – Customization requirements can determine model eligibility.

Question 25

The security architect at School of Fine Art is preparing a recommendation for a personalization program. The recommendation must cap response length to control latency and output-token cost. 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 7 business workflows and 173 recent production cases.

1. Lower temperature
2. Use prompt caching when supported and repeated context is large
3. Higher temperature
4. Shorter maximum output
5. Reduce unnecessary input tokens

Correct answer: D

Why: Limiting generated length can reduce both wait time and token spend. It directly addresses the requirement in this scenario.

Option review:

A: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Higher temperature generally increases randomness and variation. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Limiting generated length can reduce both wait time and token spend. It directly addresses the requirement in this scenario.

E: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Shorter maximum output – Limiting generated length can reduce both wait time and token spend.

Question 26

Northwind Analytics has completed discovery for a developer-productivity pilot. Before implementation, the risk manager must decide how to return supporting source passages with an answer so users can verify where the information came from. 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 210 internal users across 4 business units.

1. Amazon Bedrock Knowledge Bases
2. Retrieval Augmented Generation (RAG)
3. RAG grounding
4. Update the knowledge source instead of fine-tuning
5. RAG citation and grounding

Correct answer: E

Why: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. It directly addresses the requirement in this scenario.

Option review:

A: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: RAG retrieves relevant content at request time and adds it to the model context. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Grounding with RAG can reduce unsupported generation when retrieval sources are relevant and trustworthy. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: RAG is a strong fit for dynamic knowledge that must be refreshed independently of the model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. It directly addresses the requirement in this scenario.

Learning point: RAG citation and grounding – Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable.

Question 27

While planning a fraud-review pilot, Litware Financial identifies this requirement: prioritize a model that meets quality requirements at lower expected inference spend. Which option should the security architect prioritize if the goal is to keep the design easy to explain? Existing application interfaces can accommodate any of the listed choices, so functional fit is the deciding factor. The workload processes about 247 requests during its busiest hour and has a documented fallback path.

1. Lower-cost model
2. Smaller model
3. Model complexity
4. Cost
5. Customizable model

Correct answer: A

Why: Cost is a first-class model-selection criterion once minimum quality and governance gates are met. It directly addresses the requirement in this scenario.

Option review:

A: Cost is a first-class model-selection criterion once minimum quality and governance gates are met. It directly addresses the requirement in this scenario.

B: Smaller models can improve economics and latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Complexity can raise cost, latency, and operational burden. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Model economics should be evaluated at expected production scale. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Customization requirements can determine model eligibility. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Lower-cost model – Cost is a first-class model-selection criterion once minimum quality and governance gates are met.

Question 28

A proof of concept at A. Datum Research exposed a design decision for the risk manager: the solution must increase output diversity and creativity when some variability is acceptable. 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 284 representative records from 6 approved data sources.

1. Limit output length when long responses are not needed
2. Higher temperature
3. Shorter input context
4. Account for both input and output token rates
5. Lower temperature

Correct answer: B

Why: Higher temperature generally increases randomness and variation. It directly addresses the requirement in this scenario.

Option review:

A: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Higher temperature generally increases randomness and variation. It directly addresses the requirement in this scenario.

C: Reducing unnecessary input can improve efficiency and sometimes quality. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Lower temperature generally reduces randomness in token selection. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Higher temperature – Higher temperature generally increases randomness and variation.

Question 29

Coho Winery is documenting the target state for a compliance-assistant prototype. The security architect needs a solution that can implement managed retrieval for Bedrock applications over configured enterprise data sources. Which option is the strongest fit? The review committee wants a direct mapping from the requirement to the chosen capability. The first release supports 3 departments and is reviewed every 321 days.

1. Output validation
2. Tool-result injection
3. Amazon Bedrock Knowledge Bases
4. RAG citation and grounding
5. Confidence scoring

Correct answer: C

Why: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. It directly addresses the requirement in this scenario.

Option review:

A: Validation catches outputs that are inconsistent with authoritative data or format requirements. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Agentic applications use tool outputs as context so the model can continue from authoritative external information. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG. It directly addresses the requirement in this scenario.

D: Grounded retrieval can improve traceability and reduce unsupported generation when sources are reliable. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Confidence signals can support routing and human review, but they must be validated for the actual application. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Amazon Bedrock Knowledge Bases – Knowledge Bases for Amazon Bedrock can manage ingestion, retrieval, and grounding workflows for RAG.

Question 30

Lucerne Retail is reviewing a forecasting initiative. The risk manager has one primary requirement: select a model that can natively accept the required combination of text and images. 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 358-millisecond internal response target for the affected workflow.

1. Multilingual model
2. Customizable model
3. Low-latency model
4. Multimodal model
5. Context-window constraints

Correct answer: D

Why: Modality support must match application inputs and outputs. It directly addresses the requirement in this scenario.

Option review:

A: Language coverage and quality should be tested for the actual user population. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Customization requirements can determine model eligibility. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Latency requirements can outweigh marginal benchmark gains from a larger model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Modality support must match application inputs and outputs. It directly addresses the requirement in this scenario.

E: Input and output limits can constrain application design. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Multimodal model – Modality support must match application inputs and outputs.