AWS Certified Generative AI Developer - Professional Practice Test
The AWS Certified Generative AI Developer - Professional practice page covers building generative AI applications on AWS, with emphasis on prompt design, retrieval-grounded responses, output evaluation, and responsible handling of user data. The question bank also explores practical decisions around model changes, application architecture, and reducing unsupported or unsafe outputs.
Exam overview
About the AWS Certified Generative AI Developer - Professional Exam
Generative AI development requires more than selecting a model: developers must shape inputs, connect relevant context, test behavior, and manage changes over time. Practice scenarios address instructions that constrain answers to supplied information and retrieval-based designs that ground responses in company content. They also touch on evaluating generated outputs against reference answers, recording prompt and model versions alongside results, and comparing candidate models before deployment. Other scenarios examine separating application logic from a particular model provider, screening unsafe input, and reducing exposure of sensitive transcript data before indexing. These decisions support traceable, maintainable applications whose behavior can be assessed as prompts, models, and data evolve. The examples provide a practical way to reason through design tradeoffs rather than assume one model or configuration fits every workload.
Sample Practice Questions
Review a selection from this exam's published US Practice Tests practice bank.
A team is building an assistant that must summarize meeting notes but should not invent details that were not discussed. Which instruction is most useful in the prompt?
Show answer
Correct answer: AExplicitly instructing the model to acknowledge missing information helps reduce hallucination and keeps the summary faithful to the source notes. This is a practical guardrail for grounded summarization tasks.
A developer is building a document Q&A application on AWS and wants the model to answer using retrieved company content while minimizing unsupported claims. Which design choice best helps with this goal?
Show answer
Correct answer: AGrounding the model with retrieved source passages and explicitly constraining it to use only that context is a standard way to reduce unsupported or invented answers. This is a foundational retrieval-augmented generation practice, not a guarantee, but it usually improves factual alignment for document-based Q&A.
A team wants to measure whether model outputs match a reference answer set during testing. Which evaluation approach is most suitable?
Show answer
Correct answer: AA labeled gold-standard set provides a direct basis for judging whether model outputs align with expected answers.
A product team wants to keep a record of prompt changes, model versions, and evaluation results so they can trace why output behavior changed. What is the best practice?
Show answer
Correct answer: AVersioning prompts, models, and evaluation results makes it possible to trace regressions, compare experiments, and reproduce behavior. This is important for operational control of generative AI systems.
A team notices that a model’s style and wording change after an update, even though the prompt is unchanged. What is the best practice to manage this over time?
Show answer
Correct answer: BKeeping versioned records of prompts, models, and test outcomes makes it possible to trace why behavior changed. This supports debugging, rollback decisions, and controlled experimentation.
A company wants to compare a new model version against the current one before rolling it out to users. Which practice is the best fit?
Show answer
Correct answer: AComparing the new and current versions on the same evaluation set helps isolate behavior changes and supports safer rollout decisions.
A team wants to keep application logic independent from the specific model provider so they can change models later with minimal code changes. What design pattern is most appropriate?
Show answer
Correct answer: BAn abstraction layer separates business logic from provider-specific details, making it easier to swap models or services later with less refactoring.
How to use this practice test
Access is free. When questions are available, a session draws up to 20 random questions from the 50-question practice bank. Review the results after each session to spot gaps in areas such as retrieval grounding, evaluation, version tracking, and data privacy, then revisit those decisions before another attempt.
Preparation tips
- For retrieval-based Q&A, distinguish the roles of document indexing, retrieval context, and prompt instructions in keeping answers grounded.
- Practice choosing evaluation methods that compare generated responses with a reference set, and consider what the results reveal about output quality.
- Track prompts, model versions, and evaluation outcomes together so changes in wording or behavior can be investigated and compared.
- For transcript-based applications, identify sensitive information that should be removed or masked before content is indexed for retrieval.
- Compare candidate models before rollout, and consider how an abstraction layer can limit application changes when switching providers.