AWS AIP-C01 sample questions with answers

AWS AIP-C01 sample questions

AWS Certified Generative AI Developer – Professional, Amazon Web Services (AWS).

1. Question 1

   Domain: Foundation Model Integration, Data Management, and Compliance

   A company wants to switch the model ID and provider used by its Lambda-based GenAI API without redeploying code, and wants to roll back instantly if quality drops.

   Which approach meets these requirements?

   1. Hard-code a list of model IDs in the function and redeploy through the pipeline whenever the choice changes
   2. Keep the model choice in AWS AppConfig, read it at runtime, and roll out with a deployment strategy
   3. Store the model ID in a Lambda environment variable and publish a new function version for each change
   4. Put the model ID in the API Gateway stage name so that clients choose a model by calling a different URL
   Show the answer

   Answer: B. Keep the model choice in AWS AppConfig, read it at runtime, and roll out with a deployment strategy

   AWS AppConfig changes application behavior in production without redeploying code, with validators and gradual deployment strategies, and lets you roll back quickly if issues occur.

   Checked against: https://docs.aws.amazon.com/appconfig/latest/userguide/what-is-appconfig.html

2. Question 2

   Domain: Foundation Model Integration, Data Management, and Compliance

   A team changes prompts several times a week and has twice shipped a change that silently broke an output format that a downstream parser relies on.

   Which practice prevents this most reliably?

   1. Version and review prompts like code, and run a regression suite over saved cases before shipping
   2. Raise the temperature so the model explores more formats and can recover from a broken prompt
   3. Let each developer edit the production prompt directly, so fixes reach users as fast as possible
   4. Add the words "be careful with the format" to the prompt whenever a format bug is reported
   Show the answer

   Answer: A. Version and review prompts like code, and run a regression suite over saved cases before shipping

   Prompts as code means prompts are versioned, reviewed and covered by regression tests run against recorded cases, so a format regression fails a test before it reaches production.

3. Question 3

   Domain: Foundation Model Integration, Data Management, and Compliance

   An architect is choosing between a fast, low-cost model and the most capable model for a new summarization feature, and has an evaluation set ready.

   Which model-selection approach is recommended?

   1. Always start with the newest and largest model because cost can be optimized later without re-testing
   2. Pick the model with the highest public leaderboard score, since benchmarks predict task performance
   3. Start with a fast, cheap model and upgrade only where evaluations show a quality gap
   4. Choose the model with the largest context window, since summarization quality depends only on context size
   Show the answer

   Answer: C. Start with a fast, cheap model and upgrade only where evaluations show a quality gap

   A common starting point is to begin with a fast, cheap model and upgrade only where your own evals show a gap (or start high and move down once prompts are stable). Either way, decisions are driven by task-specific evals, not leaderboards.

4. Question 4

   Domain: Implementation and Integration

   A developer is adding an action group to an Amazon Bedrock agent.

   Which schemas can define the parameters the agent elicits for that action group?

   1. A JSON Schema stored as a Prompt Management variable, or an AWS CloudFormation template parameter section
   2. A Protocol Buffers definition, or a GraphQL schema hosted in AWS AppSync for the agent to read
   3. An AWS Glue Data Catalog table definition, or an Amazon DynamoDB key schema for the user table
   4. An OpenAPI schema or a function detail schema
   Show the answer

   Answer: D. An OpenAPI schema or a function detail schema

   Each action group uses either an OpenAPI schema, which defines API operations and their parameters, or a function detail schema that defines the parameters the agent can elicit.

   Checked against: https://docs.aws.amazon.com/bedrock/latest/userguide/agents-how.html

5. Question 5

   Domain: Implementation and Integration

   Developers want an AI assistant in their IDE that can chat about code, complete code inline, generate new code and scan for security vulnerabilities.

   Which AWS service provides this?

   1. Amazon Q Business
   2. Amazon CodeGuru Profiler
   3. Amazon Q Developer
   4. AWS Cloud9
   Show the answer

   Answer: C. Amazon Q Developer

   Amazon Q Developer provides software development assistance in the IDE: chat about code, inline completions, new code generation, security scanning and code upgrades.

   Checked against: https://docs.aws.amazon.com/amazonq/latest/qdeveloper-ug/what-is.html

6. Question 6

   Domain: Implementation and Integration

   When a CRM system records a new support case, a summary should be generated and posted back to the case, without the CRM calling the model directly.

   Which integration pattern gives loose coupling?

   1. Give the CRM database read access to Bedrock so that the model polls the case table for changes
   2. Publish the case event to an Amazon EventBridge event bus and have a rule invoke a Lambda target that calls Amazon Bedrock
   3. Have the CRM call InvokeModel synchronously inside its save transaction and wait for the summary
   4. Copy the CRM database to Amazon S3 every hour and summarize all of the cases in one batch job
   Show the answer

   Answer: B. Publish the case event to an Amazon EventBridge event bus and have a rule invoke a Lambda target that calls Amazon Bedrock

   Amazon EventBridge receives events and routes them to targets such as Lambda, so producers and consumers are decoupled; the Lambda target can call Bedrock and write the summary back.

   Checked against: https://docs.aws.amazon.com/eventbridge/latest/userguide/eb-what-is.html

7. Question 7

   Domain: AI Safety, Security, and Governance

   A pre-processing Lambda function must find and redact PII in each incoming chat message (a few kilobytes) before it is stored or sent to a model.

   Which Amazon Comprehend mode fits?

   1. Real-time PII analysis on each single document
   2. An asynchronous batch PII job started once a night
   3. Custom entity recognizer training per message
   4. Topic modeling over the chat history
   Show the answer

   Answer: A. Real-time PII analysis on each single document

   Comprehend PII detection can locate or redact PII. Real-time analysis handles a single document (up to 100 KB of UTF-8 text) per call, and asynchronous batch jobs handle collections of documents.

   Checked against: https://docs.aws.amazon.com/comprehend/latest/dg/how-pii.html

8. Question 8

   Domain: AI Safety, Security, and Governance

   Some evaluation criteria for a legal drafting assistant, such as tone and appropriateness for the firm's clients, need judgment from the firm's own lawyers.

   Which Amazon Bedrock evaluation type fits?

   1. A programmatic evaluation job that uses a built-in dataset only
   2. A retrieve-only RAG evaluation job with a context relevance metric
   3. A latency benchmark that uses the Priority service tier on each model
   4. A model evaluation job that uses human workers from your work team
   Show the answer

   Answer: D. A model evaluation job that uses human workers from your work team

   Model evaluation jobs that use human workers bring in ratings from your employees or subject-matter experts, through a work team you manage.

   Checked against: https://docs.aws.amazon.com/bedrock/latest/userguide/evaluation.html

9. Question 9

   Domain: Operational Efficiency and Optimization for GenAI Applications

   Several departments share one foundation model on on-demand inference. Finance wants each department's token spend broken out, and engineering wants per-department usage metrics.

   Which approach serves both needs?

   1. Create a separate AWS account for each department and buy Provisioned Throughput in each one
   2. Use a different temperature for each department and group invocation logs by temperature
   3. Invoke through tagged application inference profiles, one per department
   4. Put the department name in the system prompt and count the tokens in a nightly Lambda job
   Show the answer

   Answer: C. Invoke through tagged application inference profiles, one per department

   Application inference profiles let you attach cost allocation tags to on-demand invocations and collect usage metrics for the requests sent through each profile.

   Checked against: https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles.html

10. Question 10

    Domain: Testing, Validation, and Troubleshooting

    Users report that an assistant still quotes a pricing policy that was replaced in the S3 data source two days ago.

    What should the team check first?

    1. Whether a data source sync ran after the update, since changed files must be synced to be re-indexed
    2. Whether the generator model's temperature is too high, because high temperature brings back old facts
    3. Whether the reranker is active, because rerankers prefer older documents that have more citations
    4. Whether the S3 bucket has versioning on, because versioned buckets always serve old object versions
    Show the answer

    Answer: A. Whether a data source sync ran after the update, since changed files must be synced to be re-indexed

    Each time files are added, modified or removed in a data source, the data source must be synced to re-index them into the knowledge base; until then retrieval returns the old content.

    Checked against: https://docs.aws.amazon.com/bedrock/latest/userguide/kb-data-source-sync-ingest.html