Choose an AWS AI credential for your role and practice with a small Bedrock or SageMaker project.

AWS Certified AI Practitioner

Start here if you evaluate AI use cases, explain service choices, or help a team adopt AI responsibly. Practice identifying when an application needs retrieval, a prompt change, or a different model.
AWS Certified Machine Learning Engineer – Associate

Choose this route if you prepare data, train and deploy models, or operate ML pipelines. Practice reproducible training, model registration, and monitoring.
AWS Certified Generative AI Developer – Professional

Choose this route if you build production generative AI applications. Practice grounding answers, testing retrieval quality, and restricting tools and data access.
Start with one end-to-end deployment
Set up an isolated sandbox and a budget alert. Call a Bedrock model with a narrowly scoped IAM role, then add retrieval over a small public document set. Track which chunks were returned so you can tell a retrieval failure from a model error. Keep the prompt, model configuration, and sample questions in version control.
Practice for the kind of work you want
For AI Practitioner, explain a model or service choice to a stakeholder, including when a simpler approach fits. For ML Engineer, use SageMaker AI to train a baseline, check for data leakage, and rehearse rollback. For Generative AI Developer, test a Bedrock app for misleading citations and prompt injection. Give any tool a scoped role and require approval for writes.
Show what changed after testing
Publish a diagram, a small evaluation table, and a short runbook. Compare a baseline with the improved version using the same cases. Include one failure you fixed, one you still see, and the cost of running the test. Remove sample endpoints, indexes, and storage when finished.
Build something you can show
Build a read-only support assistant over public documents. Use Amazon Bedrock for inference, record which documents supported each answer, and test both answerable and unanswerable questions. Give the application role access to the approved corpus only. For an ML engineering portfolio, train a separate baseline model in SageMaker AI and record how you would detect drift and roll it back.