AWS Certified Generative AI Developer - Professional (AIP-C01) Practice Exam

What the AWS Certified Generative AI Developer - Professional (AIP-C01) exam covers

• Foundation Model Integration, Data Management, and Compliance240 questions
• Implementation and Integration201 questions
• AI Safety, Security, and Governance154 questions
• Operational Efficiency and Optimization for GenAI Applications95 questions
• Testing, Validation, and Troubleshooting86 questions

Free AIP-C01 practice test questions

A sample of 10 questions with answers and explanations. Sign up free to practice all 776.

1. Question 1Foundation Model Integration, Data Management, and Compliance

   A team wants to test a revised wording of a production prompt template against real traffic before fully replacing the current wording. Which practice supports this safely?

   • ADisable the guardrail configuration while the new wording is evaluated
   • BOverwrite the existing production template file with the new wording
   • CSave the revision as a new version and compare it to the current oneCorrect
   • DIncrease the model's maximum token output limit before testing begins
   ✓ Correct answer: C

   Versioning preserves the known-good current wording while the revision is evaluated separately, so the application can roll back instantly if the new wording underperforms. Overwriting the live template removes that safety net, and token limits or guardrails are unrelated controls.

   Why the other options are wrong

   • ADisabling guardrails introduces unrelated content risk and does not help compare prompt wording quality.
   • BOverwriting the production template directly removes the ability to compare against or roll back to the prior wording.
   • DRaising the output token limit does not help evaluate whether a revised wording performs better.
2. Question 2Foundation Model Integration, Data Management, and Compliance

   A data science team needs to generate summaries for two million archived documents overnight without maintaining a persistent low-latency connection. Which Amazon Bedrock feature is designed for this workload?

   • ABatch inference jobs, which process large prompt sets stored in Amazon S3Correct
   • BInvokeModelWithResponseStream, streaming out each summary as it completes
   • CThe Converse API, invoked once in a loop for each document in sequence
   • DThe Bedrock Agents runtime, used to orchestrate multi-step reasoning steps
   ✓ Correct answer: A

   Batch inference is built for high-volume, non-real-time processing, reading input records from an S3 location and writing results back to S3 without requiring individually managed synchronous requests. Looping synchronous calls or streaming responses is far less efficient for millions of records.

   Why the other options are wrong

   • BStreaming reduces perceived latency in interactive sessions, not large-scale offline batch throughput.
   • CLooping individual Converse calls for two million documents is slow and misses Bedrock's bulk processing capability.
   • DThe Agents runtime orchestrates multi-step reasoning and tool use, unrelated to bulk offline summarization.
3. Question 3Foundation Model Integration, Data Management, and Compliance

   A team is evaluating whether the ongoing Provisioned Throughput cost of hosting a fine-tuned custom model is justified for a feature with low, infrequent traffic. Which comparison is most relevant to this decision?

   • AWhether the training job's epoch count was high enough to fully justify any hosting cost
   • BWhether continued pre-training would remove the need for Provisioned Throughput entirely
   • CWhether model distillation would eliminate the requirement to label any training data
   • DWhether the reserved Provisioned Throughput cost outweighs the benefit for low trafficCorrect
   ✓ Correct answer: D

   Because Provisioned Throughput is billed for reserved capacity regardless of how much it is actually used, low and infrequent traffic may not justify its ongoing cost compared to simpler alternatives like prompt engineering or RAG on a base model, making this cost-versus-usage tradeoff the central decision factor rather than training epoch counts or unrelated technique choices.

   Why the other options are wrong

   • AEpoch count relates to training quality, not to whether ongoing hosting cost is justified by traffic volume.
   • BContinued pre-training still produces a custom model that itself requires Provisioned Throughput to host, it does not remove that requirement.
   • CDistillation still requires teacher-generated training data and produces a model that itself needs hosting capacity, it does not remove the labeling need entirely.
4. Question 4Foundation Model Integration, Data Management, and Compliance

   A knowledge base source bucket contains many near-duplicate copies of the same policy document, differing only in minor formatting from repeated exports. What problem does this most directly cause for retrieval quality?

   • ARetrieved results become dominated by redundant near-duplicate chunksCorrect
   • BThe vector store automatically merges duplicates into one higher-quality entry
   • CThe embedding model refuses to process duplicates and returns an error
   • DIngestion jobs fail whenever two chunks share a similar embedding value
   ✓ Correct answer: A

   When many near-identical copies of the same content exist, several of them tend to score similarly high for a related query, so a fixed number of retrieved results can fill up with repeated information instead of a diverse set of relevant passages. Deduplication before ingestion helps avoid this. Embedding models and ingestion jobs do not reject or auto-merge duplicate content on their own.

   Why the other options are wrong

   • BVector stores do not automatically detect and merge near-duplicate entries during ingestion.
   • CEmbedding models process whatever text they are given; duplicate content does not cause a processing error.
   • DSimilar embedding values between chunks do not cause an ingestion job to fail.
5. Question 5Implementation and IntegrationSelect all that apply

   A developer building an async Bedrock workflow wants clients to be notified the moment a long video-analysis job finishes, without requiring the client to repeatedly ask the server whether it is done. Which two actions together accomplish this goal? (Choose two.)

   • AThe server deletes the job record immediately after accepting the initial submission
   • BThe client keeps a single connection open and blocks until the job eventually completes
   • CThe client sends a new request every second asking whether the job has completed yet
   • DThe server invokes the registered callback URL once the job result becomes availableCorrect
   • EThe client registers a callback URL or webhook endpoint when submitting the jobCorrect
   ✓ Correct answer: D, E

   In a callback-based pattern the client supplies a webhook or callback URL at submission time, and once the asynchronous job finishes, the server makes an outbound call to that URL delivering the result or a reference to it. This avoids both the overhead of repeated status polling and the resource cost of holding a connection open for an indeterminate duration.

   Why the other options are wrong

   • ADeleting the job record immediately would make it impossible to look up or deliver the result once processing finishes.
   • BBlocking on a single open connection for a long video-analysis job risks exceeding connection and gateway timeout limits.
   • CRepeated polling requests are exactly the pattern the client wants to avoid according to the scenario, not the solution.
6. Question 6Implementation and Integration

   When constructing a toolResult message to send back to the Converse API, why must the toolUseId field exactly match the id from the preceding toolUse block?

   • AIt lets Bedrock verify the caller's IAM identity before accepting results
   • BIt lets Bedrock decide which foundation model will handle the next turn
   • CIt lets Bedrock calculate the billing charge for that tool execution step
   • DIt lets Bedrock correlate the result with the correct pending tool callCorrect
   ✓ Correct answer: D

   Each tool invocation the model requests carries a unique toolUseId, and a mismatched or missing id in the returned toolResult causes a validation error because the API cannot tell which pending call the result answers. This matters especially when multiple tool calls are outstanding in the same turn.

   Why the other options are wrong

   • AtoolUseId is a correlation identifier for tool calls, not an authentication mechanism for IAM.
   • BModel selection is configured separately and is unrelated to the toolUseId field.
   • CBedrock billing is based on token usage, not on toolUseId values.
7. Question 7AI Safety, Security, and Governance

   A model card notes that a foundation model was primarily trained and evaluated on data from a single geographic region. A team wants to deploy the model globally for customer sentiment analysis. What consideration should this disclosure raise?

   • AInference cost will rise proportionally with distance from users
   • BThe model will automatically translate all inputs regardless of region
   • CThe disclosure is irrelevant since sentiment analysis is a simple task
   • DPerformance and fairness may vary outside the training distributionCorrect
   ✓ Correct answer: D

   Models tend to perform best on data resembling their training distribution, so a model trained mainly on one region's language patterns, idioms, and cultural context may misinterpret sentiment expressed differently elsewhere, producing inconsistent or unfair results for global users. This is exactly the kind of limitation model cards are meant to surface for evaluation before broad deployment.

   Why the other options are wrong

   • AInference cost relates to compute and infrastructure, not the geographic composition of the training data.
   • BModel cards documenting training data scope say nothing about built-in automatic translation capability.
   • CSentiment analysis is sensitive to cultural and linguistic nuance, so the regional training disclosure is directly relevant.
8. Question 8AI Safety, Security, and Governance

   A team configuring Amazon Bedrock model invocation logging wants both near-real-time alerting on recent invocations and cost-effective long-term storage for later analytics. What should they configure?

   • ASend logs only to S3, since S3 supports real-time metric filter alarms natively
   • BSend logs only to CloudTrail, which can serve both alerting and long analytics
   • CConfigure both destinations: CloudWatch Logs for alerting, and S3 for storageCorrect
   • DSend logs only to CloudWatch Logs, which already offers low-cost long retention
   ✓ Correct answer: C

   Choosing both destinations lets a team set metric filters or alarms on the CloudWatch Logs stream for immediate visibility while relying on S3, often with lifecycle policies, for cheaper long-term storage and downstream analytics with tools like Athena. CloudWatch Logs is not typically the cheapest option for long-term bulk retention, S3 does not natively evaluate log content for real-time alarms, and CloudTrail does not receive Bedrock's prompt and completion content at all.

   Why the other options are wrong

   • AS3 does not natively provide real-time metric filter alarms the way CloudWatch Logs does.
   • BCloudTrail does not carry prompt and completion content, so it cannot serve the analytics need described.
   • DCloudWatch Logs long-term storage is generally costlier than S3 for bulk retention at scale.
9. Question 9Operational Efficiency and Optimization for GenAI Applications

   A payments fraud-detection service must invoke a Bedrock model with strict, predictable response times at a high and steady request volume around the clock. Relying solely on on-demand invocation has produced inconsistent latency during shared peak usage windows. Which change most directly resolves this inconsistency?

   • AAdd more retry attempts with longer exponential backoff delays
   • BTurn on streaming for each fraud detection invocation instead
   • CReduce the prompt length sent with each fraud detection request
   • DPurchase Provisioned Throughput for the steady request volumeCorrect
   ✓ Correct answer: D

   Because on-demand capacity is shared across customers, contention during peak usage windows can produce inconsistent latency; purchasing Provisioned Throughput sized for the steady, high request volume reserves dedicated capacity that is not subject to that shared contention. This directly delivers the strict, predictable response times the service requires.

   Why the other options are wrong

   • AMore retries with longer delays would only add latency during throttling rather than eliminate the contention causing it.
   • BStreaming affects how tokens are delivered to the client, not the underlying capacity contention causing inconsistent latency.
   • CShortening prompts may modestly reduce processing time but does not resolve capacity contention causing inconsistent latency.
10. Question 10Testing, Validation, and Troubleshooting

    Two RAG evaluation dimensions, faithfulness and answer relevance, are sometimes confused. Which statement correctly distinguishes them?

    • AFaithfulness and relevance both measure only how toxic the final answer is
    • BFaithfulness and relevance both measure only the speed of the generation pipeline
    • CFaithfulness checks grounding in the context; relevance checks the query fitCorrect
    • DFaithfulness checks the query fit; relevance checks grounding in the context
    ✓ Correct answer: C

    The two dimensions are independent: an answer can be fully grounded in the retrieved passages yet fail to address what the user asked, or it can directly address the question while including ungrounded, hallucinated details.

    Why the other options are wrong

    • ANeither dimension measures toxic language.
    • BNeither dimension measures pipeline speed.
    • DThis reverses the two definitions; faithfulness is about grounding, relevance is about addressing the query.

Who this AWS Certified Generative AI Developer - Professional (AIP-C01) practice exam is for

This practice set is for anyone preparing for the AWS Certified Generative AI Developer - Professional (AIP-C01) exam at the advanced level - from first-time candidates building a foundation to experienced AWS practitioners doing a final review before test day. If you learn best by working through realistic questions and reading why each answer is right or wrong, it is built for you.

How to use this AWS Certified Generative AI Developer - Professional (AIP-C01) practice exam

1. Start with the free sample questions above to gauge your current baseline.
2. Read the full explanation on every question, including why each wrong option is wrong.
3. Track your weak domains and focus your study where you are losing the most marks.
4. Once you are scoring consistently well, take a timed, full-length mock exam.
5. Use your readiness score to decide when you are ready to book the real AWS Certified Generative AI Developer - Professional (AIP-C01) exam.

Related AWS resources

• AWS Certified Generative AI Developer - Professional (AIP-C01) study guideKey concepts
• AWS practice examsAll AWS
• Certification pathWhere this fits
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• How these questions are written and reviewedMethodology
• Report a problem with a questionCorrections
• AWS CLF-C02 practice examRelated
• AWS DOP-C02 practice examRelated
• AWS DVA-C02 practice examRelated

AWS Certified Generative AI Developer - Professional (AIP-C01) practice exam FAQ