Use the exam guide as the source of truth
Start with the current AWS exam guide and convert each domain into a checklist of decisions you can explain. Product names are not enough: practice choosing a service from a business requirement, identifying why alternatives are weaker, and describing the data, security, quality, and cost implications.
PrepKloud questions are original learning scenarios. They diagnose understanding but do not reproduce or predict live exam items. Recheck the official guide before scheduling because domains and services can change.
Build a service decision map
Separate purpose-built AI services, Amazon Bedrock foundation-model applications, and SageMaker AI custom machine-learning workflows. For each task, record input modality, desired output, customization, explainability, latency, data controls, human review, and the no-AI baseline.
- Vision and document tasks: compare Rekognition and Textract
- Language and speech tasks: distinguish Comprehend, Translate, Transcribe, Polly, and Lex
- Generative tasks: understand Bedrock models, knowledge bases, agents, and guardrails
- Custom ML: understand SageMaker data, training, deployment, monitoring, and explainability concepts
Practice responsible decisions
Responsible AI questions are rarely solved by a single filter. Connect fairness, explainability, privacy, safety, transparency, robustness, security, and human oversight to the complete lifecycle. Know what data is used, who can access it, how performance is evaluated, which failures matter, and when the system must defer to a person.
Run a four-week learning loop
Week one: AI, ML, deep learning, training and inference, common task types, and evaluation basics. Week two: AWS purpose-built AI services and service-selection scenarios. Week three: generative AI, foundation models, prompting, embeddings, RAG, Bedrock, and model adaptation. Week four: responsible AI, security, governance, mixed practice, and one project.
After each quiz, group errors by decision pattern rather than memorizing the correct option. Revisit the official source, explain the decision aloud, and test the concept in a small synthetic lab.
Choose the right AWS AI layer
| Need | Likely layer | Decision question |
|---|---|---|
| Ready-made vision, language, speech, or document capability | Purpose-built AWS AI service | Does an API already solve the bounded task? |
| Foundation-model generation or RAG | Amazon Bedrock | Do you need model choice and managed generative-AI capabilities? |
| Custom model training and ML lifecycle control | Amazon SageMaker AI | Do you need to build, tune, deploy, or monitor a custom ML model? |
| Stable rules or calculations | Deterministic software | Would AI add uncertainty without useful flexibility? |
Practical checklist
- Official exam guide reviewed this week
- Every domain mapped to decisions and evidence
- Service map tested with scenarios
- Weak areas grouped and remediated
- At least one responsible-AI project completed
- Current exam status, price, language, and policy verified on AWS
First-party sources
- AWS Certified AI Practitioner
- Amazon Bedrock User Guide
- Amazon SageMaker AI Developer Guide
- AWS Responsible AI
Source status last checked 2026-09-11. Links can change after publication.