CCNA Troubleshooting Questions

MCQhard

A developer is using Amazon Bedrock Guardrails to filter out hate speech and violence in a chatbot application. During testing, the developer notices that some prompts containing subtle hate speech are not being blocked. The developer wants to improve the detection rate without increasing false positives on benign content. Which action should the developer take?

A.Switch the Guardrails content filter to use a different foundation model for moderation.

B.Increase the Guardrails filter strength for hate speech to the highest setting.

C.Disable the hate speech filter and rely on the violence filter to catch all problematic content.

D.Add custom words or phrases to a Guardrails word filter that specifically target the subtle hate speech patterns.

AnswerD

Guardrails support custom word filters that can be configured to block specific terms or phrases. By adding the subtle hate speech patterns to a custom filter, the developer can increase detection without broadly raising sensitivity, thus minimizing false positives on benign content. This is a targeted approach.

Why this answer

To improve detection of subtle hate speech without increasing false positives, the developer should use Guardrails' custom word filters. These allow adding specific terms or phrases that the standard classifiers might miss. By targeting the exact patterns, the developer can enhance detection while keeping the broader filters at balanced settings, thus avoiding over-blocking benign content.

Exam trap

The trap here is thinking that raising the filter strength is the only way to catch more hate speech, when custom word filters offer a more precise solution.

Practice this question →

Multi-Selecteasy

Which THREE of the following are valid components of an Amazon Bedrock Agent test plan?

Select 3 answers

A.Validating the reasoning path of the agent.

B.Verifying the S3 bucket encryption.

C.Testing the accuracy of tool invocations (Action Groups).

D.Measuring model training time.

E.Ensuring Guardrails are applied to both input and output.

AnswersA, C, E

Validating the reasoning path verifies the agent's chain-of-thought and orchestration logic, not just the final response. This satisfies the test plan requirement to assess how the agent reaches conclusions, exposing flawed planning or prompt interpretation before production deployment.

Why this answer

Testing reasoning, tool invocation accuracy, and safety constraints are critical components of an agent test plan.

Practice this question →

MCQhard

A developer is troubleshooting a generative AI application that calls Amazon Bedrock. Intermittently, invocations fail with a ThrottlingException. CloudWatch metrics show that the account's on-demand quota for the model is being exceeded during peak hours. The developer wants to reduce throttling without changing the model or application logic. Which solution is most appropriate?

A.Reduce the maxTokenCount parameter to 1 to lower token consumption per request.

B.Switch the application to use a different AWS Region where the model is available and retry all requests there.

C.Implement exponential backoff with jitter in the AWS SDK retry configuration and request a quota increase for the model.

D.Enable provisioned throughput for the model and remove all retry logic from the application.

AnswerC

ThrottlingException occurs when request rate or token throughput exceeds the account's quota. Exponential backoff with jitter lets retries succeed when the limit is transient, while a quota increase raises the ceiling for sustained peak load. Together they address both the immediate retry behavior and the underlying capacity limit without altering the model or application logic.

Why this answer

Intermittent ThrottlingException from Bedrock indicates the account is exceeding its on-demand quota. Exponential backoff with jitter allows retries to succeed when throttling is transient, and requesting a quota increase raises the sustained capacity limit. This combination reduces throttling without modifying the model or application logic, addressing both immediate and long-term needs.

Exam trap

The trap here is assuming that reducing token limits or switching Regions solves throttling, when the actual causes are request rate and quota ceilings that require retry strategy and quota adjustments.

Practice this question →

MCQmedium

A developer is troubleshooting an Amazon Bedrock Agent that fails to complete a multi-step task. CloudWatch Logs show that the agent successfully invokes the action group's Lambda function, but the agent then replies to the user that it cannot proceed. The Lambda function returns a JSON object with a top-level key "result" instead of the key the agent expects. Which change will most likely resolve the issue?

A.Enable the agent's "return control" option so the agent can bypass the Lambda response and generate the final answer directly.

B.Add a new action group to the agent that duplicates the same Lambda function but uses a different API schema version.

C.Modify the Lambda function to return the response in the format defined by the action group's API schema, including the required response body structure.

D.Increase the Lambda function's memory size to 1024 MB and extend the timeout to 5 minutes to allow the agent more time to parse the response.

AnswerC

Amazon Bedrock Agents parse the Lambda response according to the OpenAPI schema attached to the action group. If the function returns an unexpected top-level key, the agent treats the invocation as malformed and stops the orchestration. Aligning the Lambda output with the schema's response body structure restores successful parsing and lets the agent continue the multi-step task.

Why this answer

Bedrock Agents rely on the action group's OpenAPI schema to validate and parse the Lambda response. When the function returns a payload whose structure does not match the schema, the agent cannot extract the expected fields and aborts the orchestration, even though the invocation itself succeeded. Correcting the Lambda to emit the schema-defined response body is the direct fix.

Exam trap

The trap here is assuming that a successful Lambda invocation in CloudWatch Logs means the agent received a usable response, when the failure is actually a schema mismatch in the returned JSON.

Practice this question →

MCQmedium

A generative AI application deployed on AWS Lambda invokes an Amazon Bedrock model and occasionally receives a ThrottlingException with the message 'Too many requests'. The developer has already confirmed that the Lambda function's reserved concurrency is well below the account limit. The application must remain reliable without over-provisioning. Which action should the developer take to handle these throttling errors most effectively?

A.Implement an exponential backoff and jitter retry strategy using the AWS SDK for the Bedrock InvokeModel API call.

B.Configure the Lambda function to use provisioned concurrency to ensure immediate scaling.

C.Increase the Lambda function's memory size to improve CPU performance and reduce invocation time.

D.Switch the model invocation to use the Amazon Bedrock Converse API instead of InvokeModel.

AnswerA

Bedrock throttling is transient and retrying with exponential backoff and jitter is the AWS-recommended approach. The SDK's built-in retry mechanism can be customized, and adding jitter prevents synchronized retries from multiple Lambda instances. This directly addresses the throttling without changing quotas.

Why this answer

ThrottlingException from Amazon Bedrock indicates that the request rate exceeded the account's quota. The correct approach is to implement retries with exponential backoff and jitter, which is a standard resilience pattern for transient errors. This allows the application to recover gracefully without manual quota increases, and it works well in distributed systems like Lambda.

Exam trap

The trap here is assuming that Lambda concurrency or memory settings control Bedrock throttling, when throttling is enforced by the Bedrock service based on request rates.

Practice this question →

MCQmedium

A developer is using Amazon Bedrock to build a customer support chatbot. The model is hallucinating frequently when retrieving internal company knowledge. Which evaluation methodology should the developer implement to quantify the ground truth accuracy of the RAG pipeline?

A.Implement a RAGAS-based evaluation pipeline to measure faithfulness and answer relevance.

B.Increase the temperature setting to 1.0 to encourage more creative responses.

C.Use Amazon CloudWatch logs to calculate the latency of every inference call.

D.Run a standard BLEU score evaluation against the model's training data.

AnswerA

RAGAS quantifies retrieval-augmented generation quality by scoring faithfulness (whether claims are entailed by retrieved context) and answer relevance, directly measuring ground-truth accuracy against the internal knowledge base. This satisfies the stem's requirement to quantify hallucination reduction in the RAG pipeline.

Why this answer

RAGAS (Retrieval-Augmented Generation Assessment) is a framework specifically designed to evaluate RAG pipelines. It measures faithfulness (whether the answer is grounded in the retrieved context) and answer relevance (how relevant the answer is to the question). These metrics directly quantify ground truth accuracy and help identify hallucinations.

Therefore, implementing a RAGAS-based evaluation is the correct approach.

Exam trap

AIP-C01 often tests evaluation methodologies for RAG, and candidates may confuse RAGAS with generic metrics like BLEU or think that increasing temperature improves accuracy.

How to eliminate wrong answers

Option B is wrong because increasing temperature to 1.0 makes the model more creative and likely to hallucinate more, not less. Option C is wrong because CloudWatch logs measure latency, not accuracy or hallucination. Option D is wrong because BLEU score evaluates text similarity against reference translations, not RAG ground truth accuracy, and it is not suitable for open-ended chatbot responses.

Practice this question →

Multi-Selecthard

A developer is troubleshooting an Amazon Bedrock Agent that is frequently hitting errors. Which TWO of the following steps should be taken to isolate the issue?

Select 2 answers

A.Restart the Bedrock service.

B.Increase the model's temperature to 1.0.

C.Check CloudWatch Logs for the associated Action Group Lambda functions.

D.Review the Agent's 'Trace' output to identify the failure point.

E.Re-train the underlying base model.

AnswersC, D

Action Group invocations run in Lambda, so CloudWatch Logs capture the request payloads, errors and timeouts that surface as agent failures. Reviewing these logs isolates whether the fault lies in the Lambda function rather than the agent's orchestration or prompt configuration.

Why this answer

Option C is correct because Bedrock Agents invoke Action Group Lambda functions to fulfill API operations, and their execution logs, errors, and timeouts are captured in CloudWatch Logs, making them the primary place to isolate backend failures. Option D is correct because the Agent's Trace output shows the step-by-step orchestration (pre-processing, orchestration, knowledge base/action group invocation, and post-processing), revealing exactly which stage or invocation failed. Option A is wrong because Bedrock is a fully managed service with no customer-accessible restart operation, and restarting would not diagnose the root cause.

Option B is wrong because raising temperature to 1.0 only increases response randomness and would not isolate an error. Option E is wrong because base foundation models cannot be retrained by customers; customization is done via fine-tuning or continued pre-training, which is unrelated to troubleshooting.

Exam trap

AIP-C01 often tests whether candidates understand that Bedrock is a managed service where you cannot 'restart' or 'retrain' base models, steering them toward the two observability tools — Lambda CloudWatch Logs and Agent Trace — that actually reveal the failure point.

Practice this question →

Multi-Selectmedium

A developer is creating a test suite for an LLM application. Which THREE of the following metrics are commonly used to evaluate RAG system performance?

Select 3 answers

A.Answer Relevance

B.Context Precision

C.Model Training Loss

D.Faithfulness

E.Inference Latency

AnswersA, B, D

Answer Relevance scores how directly the generated response addresses the user's original question, independent of factual grounding. In a RAG test suite it catches responses that are faithful to retrieved context yet drift off-topic, satisfying the need to evaluate end-to-end answer quality.

Why this answer

Faithfulness, answer relevance, and context precision are the key components of RAGAS evaluation.

Practice this question →

MCQhard

A developer is testing a retrieval-augmented generation (RAG) application built on Amazon Bedrock Knowledge Bases. For a specific question, the model returns a confident but incorrect answer, and the retrieved chunks shown in the test console are only loosely related to the question. The developer wants to diagnose whether the retrieval step or the generation step is at fault. Which approach is most effective?

A.Use the Retrieve API to fetch chunks for the question and inspect their relevance scores and content independently of generation.

B.Add a guardrail that blocks unsupported answers and observe whether the response is filtered, indicating a generation problem.

C.Switch the knowledge base to a different embedding model and re-ingest all documents, then compare the final answers.

D.Increase the model's temperature to 1.0 and re-run the query to see if the answer changes, then compare outputs.

AnswerA

The Retrieve API returns the chunks and relevance scores without invoking the generation model. By examining whether the top-ranked chunks actually contain the answer, you can determine if retrieval is failing. If relevant chunks are present but the answer is wrong, the fault lies in generation; if not, retrieval needs tuning.

Why this answer

To separate retrieval failures from generation failures, you must inspect the retrieved chunks independently. The Retrieve API returns the chunks and relevance scores without invoking the generation model, so you can verify whether the correct context was available. If the right context is present but the answer is wrong, generation is at fault; otherwise, retrieval tuning is needed.

Exam trap

The trap here is modifying generation parameters such as temperature or adding guardrails, which change output behavior without revealing whether the retrieved context was relevant.

Practice this question →

MCQhard

An application deployed on AWS Lambda is receiving 429 Too Many Requests errors when calling Amazon Bedrock. The developer has verified that account-level quotas have not been reached. What is the most likely cause?

A.The VPC endpoint policy is blocking traffic to the Bedrock runtime service.

B.The application is exceeding the tokens-per-minute (TPM) limit for the specific model in that AWS Region.

C.The Lambda function timeout is set too low for the model response time.

D.The IAM role is missing the bedrock:InvokeModelWithResponseStream permission.

AnswerB

Exceeding the model-specific tokens-per-minute quota in that Region triggers throttling independently of account-level limits, so 429s appear despite unused account capacity. Bedrock enforces per-model, per-Region TPM and RPM quotas separately, meaning a single high-volume Lambda workload can saturate the model's throughput ceiling while aggregate account usage remains low.

Why this answer

A 429 Too Many Requests from Bedrock, despite account-level quotas being fine, most commonly means the application is hitting the per-model tokens-per-minute (TPM) or requests-per-minute (RPM) limit for that specific model in that Region. Bedrock enforces model-specific quotas that are separate from account-level limits, so a single high-volume model can throttle even when overall account usage is low.

Exam trap

AIP-C01 often tests whether candidates distinguish account-level quotas from model-specific TPM/RPM quotas — a 429 with healthy account usage almost always points to the per-model limit, not IAM or networking.

How to eliminate wrong answers

Option A is wrong because a VPC endpoint policy blocking traffic would produce a connection timeout or 403, not a 429 throttling response. Option C is wrong because a Lambda timeout produces a task timeout error (or the function being killed), not a 429 from Bedrock. Option D is wrong because a missing IAM permission yields a 403 AccessDenied, not a 429.

Practice this question →

Multi-Selectmedium

A developer is validating a generative AI application that uses Amazon Bedrock. The application must produce deterministic, repeatable outputs for regression testing. Which TWO configuration changes should the developer make to achieve this? (Choose two.)

Select 2 answers

A.Enable streaming responses to capture partial outputs for comparison.

B.Increase the maxTokenCount parameter to the model's maximum to ensure complete answers.

C.Provide a fixed random seed using the model's supported seed parameter, if available for that model.

D.Set the topP parameter to 1.0 and leave temperature at its default value.

E.Set the temperature parameter to 0 in the model invocation request.

AnswersC, E

Some Bedrock models support a seed parameter that, when combined with temperature 0 or low temperature, helps produce repeatable outputs. Supplying a fixed seed reduces variability across identical requests. This is a valid additional step to improve determinism when the chosen model exposes seed support.

Why this answer

Deterministic regression testing requires minimizing sampling randomness. Setting temperature to 0 makes token selection greedy, and using a fixed seed where supported further stabilizes outputs across identical requests. These two settings together provide the most repeatable behavior available from Bedrock models, which is essential for comparing outputs across application versions.

Exam trap

The trap here is confusing parameters that control output length or delivery, such as maxTokenCount and streaming, with parameters that control randomness, such as temperature and seed.

Practice this question →

MCQeasy

A developer is testing an Amazon Bedrock application and wants to verify that the model's responses do not contain personally identifiable information (PII) before the application displays them to users. The developer needs a managed way to detect and block PII in both prompts and responses. Which AWS feature should be used?

A.Amazon Comprehend PII detection called manually after each inference to filter outputs.

B.AWS WAF with a custom rule to inspect request and response bodies for PII patterns.

C.Amazon Macie to scan the model's responses for PII before display.

D.Amazon Bedrock Guardrails with the sensitive information filter configured for PII.

AnswerD

Bedrock Guardrails provide a managed sensitive information filter that can detect and block or mask PII in both input prompts and model responses. Configuring the filter with PII entities addresses the requirement directly. It integrates with model invocation so that policy is enforced consistently without custom code.

Why this answer

Amazon Bedrock Guardrails include a sensitive information filter that can detect and block or mask PII in both input prompts and model responses. Configuring this filter enforces PII policy at inference time without custom code, satisfying the need for a managed solution that covers both directions of the interaction.

Exam trap

The trap here is choosing a general-purpose data or network security service such as Macie or WAF, which either scans data at rest or lacks PII semantics, instead of the Bedrock-native Guardrails filter.

Practice this question →