AWS AI Practitioner Cheat Sheet.
The night-before summary, built like the exam.

AWS Certified AI Practitioner — Cheat Sheet

A dense, night-before reference for the Foundational-level AWS Certified AI Practitioner exam. Skim the table, memorize the domain weights, then run the checklist.

Exam Logistics At-a-Glance

• Scheduling happens through Pearson VUE from your AWS Certification Account.
• Testing centers are independently operated facilities, not run directly by AWS.
• The 15 unscored questions are mixed in and not identifiable during the exam, so treat every question as if it counts.

Content Domains (Memorize the Weights)

1. Domain 1 — Fundamentals of AI and ML: 20%
2. Domain 2 — Fundamentals of Generative AI: 24%
3. Domain 3 — Applications of Foundation Models: 28% (largest domain — prioritize study time here)
4. Domain 4 — Guidelines for Responsible AI: 14%
5. Domain 5 — Security, Compliance, and Governance for AI Solutions: 14%

Five domains total. Domains 3 and 2 together are over half the scored exam — spend your final review hours on foundation model applications and generative AI fundamentals before circling back to the smaller domains.

Key Terms and Concepts to Memorize

• Foundation model (FM): a large pretrained model adaptable to many downstream tasks via prompting, fine-tuning, or retrieval augmentation.
• Generative AI vs. traditional ML: traditional ML predicts or classifies from labeled data; generative AI produces new content (text, image, code) from learned patterns.
• Prompt engineering: structuring inputs (instructions, context, examples) to steer model output without retraining.
• Fine-tuning vs. RAG: fine-tuning adjusts model weights on a custom dataset; retrieval-augmented generation injects external knowledge at inference time without retraining.
• Embeddings and vector databases: numeric representations of meaning used for semantic search and RAG retrieval.
• Bias, fairness, and explainability: core pillars of responsible AI — expect scenario questions about detecting and mitigating biased outputs.
• Hallucination: a model generating plausible-sounding but factually incorrect content — a recurring exam trap topic.
• Shared responsibility model: AWS secures the cloud infrastructure; customers secure their data, access configuration, and model usage.
• Guardrails and content filtering: mechanisms to constrain model outputs for safety and compliance.
• Model evaluation metrics: know the difference between metrics used for classification/regression tasks versus qualitative evaluation of generative outputs.

Common Gotchas and Traps

• Don't assume every question is scored — but don't try to guess which ones aren't, since unscored items are indistinguishable during the test.
• The compensatory scoring model means you can't "fail" a single domain outright, so don't panic over a rough patch of questions — keep moving and let the overall score average out.
• Watch for questions that test conceptual understanding rather than memorized service names — this is a Foundational exam, so expect "why would you use X" framing over deep configuration detail.
• Security and governance (Domain 5) questions often overlap with responsible AI (Domain 4) — read carefully to determine whether the question is asking about technical controls or ethical/policy guidelines.
• Time management: 90 minutes for 65 questions is roughly 83 seconds per question — flag long scenario questions and return to them rather than stalling.
• Rescheduling within 24 hours of your appointment may incur fees, so lock in logistics early.

Night-Before Checklist

• Confirm delivery method: testing center address and ID requirements, or OnVUE system check (webcam, room scan, private space) if testing online.
• Review the domain weight list once more, prioritizing Domain 3 (28%) and Domain 2 (24%).
• Skim key terms above — hallucination, RAG vs. fine-tuning, shared responsibility model, guardrails.
• Verify your AWS Certification Account login and exam confirmation details.
• Get sleep — a Foundational exam rewards clear thinking on scenario-based questions more than raw memorization.
• Remember the passing bar: 700 out of a 100–1,000 scaled score, compensatory across all domains.
• Plan arrival or online check-in time with buffer — late arrival can forfeit your slot.