IT & cloud certifications · course 53 of 105

AWS AI Practitioner: Which Service, Which Metric

Work through the AWS AI Practitioner (AIF-C01) exam guide: the AI words in AWS’s own wording, precision and recall from a confusion matrix, which AWS service does which job, and which safeguard fits.

4 modules

Try a question

Not on sale yet. When sales open: $4.99 for this course, paid once, or every course for $9.99 a month. Prices

Preview

What it teaches

The words and the decisions behind the AWS Certified AI Practitioner (AIF-C01) exam guide, in 4 modules that follow its 5 domains. M0: the AI words (AI, machine learning, deep learning, generative AI, foundation model, large language model, token, prompt engineering, RAG, inference) in AWS’s own wording. M1, AI and ML fundamentals (Domain 1, 20% of the exam): supervised, unsupervised and reinforcement learning, classification vs regression vs clustering, the five ML lifecycle phases as AWS lists them, and accuracy, precision, recall and F1 worked out from a confusion matrix. M2, generative AI on AWS (Domains 2 and 3, 24% + 28%): which AWS service does a stated job (Amazon Bedrock, Comprehend, Rekognition, Textract, Transcribe, Translate; SageMaker AI, Polly and Lex are on the card but are not answers), what temperature, Top P and response length do, and when RAG, fine-tuning or prompt engineering fits. M3, responsible AI, security and governance (Domains 4 and 5, 14% + 14%): AWS’s dimensions of responsible AI, Amazon Bedrock Guardrails, CloudTrail, Macie, Config, Artifact, the shared responsibility model and IAM (card only). The AI asks 3 questions, places you, shows a short card per module, then gives you one question at a time.

Not full exam prep (see “Not in it yet” under “What’s exact, and what isn’t”), and it can’t judge your own AI project.

Not real exam questions: written by Tutorpack from AWS’s public pages; not connected to AWS; no prices; no promise of passing. The exam guide reports a scaled score of 100-1,000 with 700 to pass and gives no conversion from raw answers on the pages read, so the course’s 10/12 final-test bar is its own, not a prediction.

How long

45 items (10 concept questions and 35 scored items). About 12-30 rounds for someone who already knows the area, about 60 for a beginner who misses a third of the items, plus a 12-item final: 2-3 sessions (from AI-simulated students, not timed with a real student). The tutor gives a one-line save code at the end of each session; paste it into a new chat to carry on.

The modules

4 modules, drilled one item at a time.

M0
The words
The AI words in AWS’s own wording: AI, machine learning, deep learning, generative AI, foundation model, large language model, token, prompt engineering, RAG, inference
M1
AI and ML fundamentals
Types of machine learning, classification vs regression vs clustering, the ML lifecycle, and accuracy, precision, recall and F1 from a confusion matrix
M2
Generative AI and foundation models on AWS
Which AWS service does the job, what temperature, Top P and response length do, and when RAG, fine-tuning or prompt engineering fits
M3
Responsible AI, security and governance
AWS’s dimensions of responsible AI, which AWS tool does what, and who is responsible under the shared responsibility model

Before it teaches, it asks

1
What is your background with AI or AWS (none, some, or already working with it), and when, roughly, do you want to take the exam (or not planned)?
2
A model must predict whether each incoming e-mail is 'spam' or 'not spam'. What kind of machine-learning problem is that?
3
An online shop wants its English product descriptions published in French and Spanish. Which purpose-built AWS machine-translation service does that job?

What’s exact, and what isn’t

Show

Straight from the course’s own notes.

Exam status (AWS's exam guide pages, read 2026-10-13): AWS’s AIF-C01 exam guide is live and no page read calls it retired. The guide’s Revisions page lists version 1.0 (March 26, 2026) and version 1.1 (April 30, 2026); this course follows the guide as it stood on 2026-10-13. The domain weights are 20% / 24% / 28% / 14% / 14% of scored content. None of the guide pages read (index, the five domains, in-scope services, revisions) names a retirement date or a successor exam; a script fails if the words retire, successor, AIF-C02, replaced by or discontinue appear on the saved pages. The guide changes: check AWS’s page for the current version before relying on this.
Quoted from AWS: Every card line, every M0 Answer and every working line that states a fact is a quote from an AWS page read 2026-10-13 (88 quotes from 28 pages). A script checks each word for word on a saved copy of its page; it does not re-fetch the live pages.
Computed by script: The 4 metric keys in M1 (exact fractions, then the 2-place rounding each question states) and the Top P Key in M2. The metric formulas and the Top P rule are AWS quotes; the numbers come from the script.
Checked by script: All 35 wrong answers. The 4 metric answers and the Top P answer recompute the slip’s own number or words; the other 30 are checked against what the AWS quotes say each name covers (the key’s name covers what the question needs, no other name in its group does, the wrong name doesn’t). The lifecycle’s five phases are checked in AWS’s order on the saved page, the confusion-matrix counts must add up to the number the question states, and each service named is checked against the exam guide’s in-scope list.
Term and service Keys are on the cards on purpose: A card names each service, term and tool next to its AWS sentence. The items ask you to apply them to a stated job. A script checks that no card prints a computed key.
Honest limits: Most questions are made-up business situations, each stating the facts that decide the answer; a few are plain questions. No open licence was found on the AWS pages read, so the course uses short quotes only and names the page. The course can’t see your AWS account or judge a real system.
Not in it yet: Exam-guide task statements 2.2 (capabilities and limits of generative AI, such as hallucinations) and 3.4 (evaluating foundation models: ROUGE, BLEU, BERTScore); agentic AI (agents, MCP, Amazon Bedrock AgentCore, Strands Agents, Kiro); Amazon Q and Amazon Quick; SageMaker Clarify, Model Monitor and Augmented AI (not named in the five current domain pages; Clarify and Amazon Q appear only in the Revisions page’s version 1.1 column); types of inferencing, data types, MLOps, token-based pricing; in-scope services with no item yet (Personalize, Nova, JumpStart, KMS, Secrets Manager, Inspector, Trusted Advisor and others); and Lex, Polly, SageMaker AI, IAM and four of the eight responsible-AI dimensions (controllability, explainability, privacy and security, governance), which are on a card but have no item of their own (Polly, SageMaker AI, IAM, explainability, privacy and security and governance appear as wrong answers).