Study the official exam domains. Each lesson teaches the concept from the vendor exam guide and first-party docs.
This page shows the complete course curriculum. Enroll now to access all video lessons, hands-on projects, and downloadable resources.
Enroll in This CourseArtificial intelligence is the umbrella. Machine learning is one way to get there. Deep learning is the machine learning method that uses neural networks with many hidden layers. A first-time reader only needs one rul...
The exam asks you to pick a learning type from the data you have and the output you want. Supervised learning needs labeled examples. Unsupervised learning finds structure in unlabeled examples. Reinforcement learning...
A model is only as useful as the file you give it. The exam asks you to name the data before you name the technique. Labeled versus unlabeled decides the learning type. Structured versus unstructured decides how much...
Training builds the model. Inference is the call that uses it. AWS also writes inferencing for the same idea. Batch inference scores a pile of records when you can wait. Real-time inference answers one request with lo...
Domain 1 asks you to recognize the job before you recognize the service. Computer vision reads images and video. Natural language processing reads and writes language. Speech turns audio into text or text into audio....
The exam does not treat every problem as a model. Task 1.2 asks you to recognize when AI or machine learning adds value, and when it does not. Value looks like help for a human decision, a process that must scale past...
Task 1.2 asks you to explain the managed AI and ML services AWS already trained for common jobs. Amazon SageMaker AI is the workshop where you build, train, and host your own models. Amazon Transcribe, Amazon Translat...
Task 1.3 asks you to walk the machine learning pipeline in order. Data collection, exploratory data analysis, preprocessing, feature engineering, training, hyperparameter tuning, evaluation, deployment, and monitoring...
The last Domain 1 task is operational. Where does the model come from. How does an application call it. How do you know the model is good. How do you know the project is worth it. Open-source and pre-trained checkpoin...
Task 2.1 starts with the raw units a generative model actually consumes. A prompt is not one blob of English. The model splits that text into tokens, turns meaning into embeddings, and compares those embeddings as vec...
Task 2.1 asks you to name the model class that made generative AI practical. A foundation model is a large deep learning network trained on a broad, mostly unlabeled corpus so it can do many general jobs from a prompt...
Task 2.1 lists prompt engineering next to tokens and foundation models because a generative model is open-ended. One word can produce a long reply. A useful reply needs instruction, context, and a stated shape. Prompt...
Task 2.1 asks for the foundation model lifecycle. Data selection, model selection, pre-training, fine-tuning, evaluation, deployment, and feedback. The expensive step is pre-training. Fine-tuning changes weights on a...
Task 2.1 pairs multi-modal models with diffusion models because not every generative job is next-token text. A multimodal model accepts or produces more than one kind of signal. Text, image, video, speech, or a mix. D...
Task 2.2 asks why a firm would use generative AI and how to pick a model. The exam's advantage words are adaptability, responsiveness, and simplicity. The selection words are model type, performance, capability, const...
Task 2.2 names four disadvantages. Hallucinations, interpretability, inaccuracy, and nondeterminism. They all come from the same engine you learned in lesson 2.2. The model samples the next token from a probability di...
Task 2.3 asks you to name the AWS service that turns foundation models into an application API. Amazon Bedrock is that service. Fully managed. You pick a model, send a prompt, and receive a completion. You do not stan...
Task 2.3 lists Amazon Q next to PartyRock, an Amazon Bedrock Playground. They sit at opposite ends of the same idea. PartyRock is how a person learns prompting and model choice in a browser with no AWS account. Amazon...
Task 2.3 names Amazon SageMaker JumpStart beside Bedrock, PartyRock, and Amazon Q. JumpStart is the machine-learning hub inside SageMaker AI. You discover, deploy, fine-tune, and evaluate foundation models, including...
Domain 3 is Applications of Foundation Models. It is 28 percent of scored AIF-C01 content. Task 3.1 starts with design. You pick a pre-trained model, then you set inference parameters so the live call matches the job....
Task 3.1 asks you to name the AWS services that store embeddings in a vector database. Lesson 2.1 already taught the units. A chunk becomes a vector. Nearby vectors mean nearby meaning. This lesson is the application...
Task 3.1 defines Retrieval Augmented Generation and asks for its business applications, including Amazon Bedrock and a knowledge base. Lesson 2.4 taught the contrast. RAG leaves weights alone. Fine-tuning changes weig...
Task 3.1 asks for the role of agents in multi-step tasks and names Agents for Amazon Bedrock. Retrieval from lesson 3.3 fetches text and then writes. An agent also acts. It breaks a request into steps, asks the user f...
Task 3.1 pairs Retrieval Augmented Generation with Amazon Bedrock and a knowledge base. Lesson 3.3 designed the four stages. This lesson is the managed product that runs them. Amazon Bedrock Knowledge Bases convert yo...
Task 3.2 is choose effective prompt engineering techniques. Lesson 2.3 taught the wrapper. Instruction, context, zero-shot, few-shot, and chain-of-thought. This lesson is application scale. You store a reusable templa...
Task 3.4 asks how you evaluate foundation model performance. Automatic scores on a benchmark, and humans on a work team. The exam names Recall-Oriented Understudy for Gisting Evaluation (ROUGE), Bilingual Evaluation U...
Task 3.1 asks for cost tradeoffs of foundation model customization and for latency as a selection criterion. Lesson 2.10 scored JumpStart versus Bedrock bills. This lesson is the production application. Every live cal...
Task 3.3 asks how you prepare data to fine-tune a foundation model. Curation, governance, size, labeling, representativeness, and reinforcement learning from human feedback (RLHF). Task 3.4 already used humans as rate...
Task 3.1 closes Domain 3 with a design choice. You already picked a model and set temperature in lesson 3.1. You already know the stores, RAG, agents, knowledge bases, prompt chains, scores, bills, and human loops. Th...
Domain 4 is Guidelines for Responsible AI. It is 14 percent of scored AIF-C01 content. Task 4.1 starts with the features of a responsible system. The exam list is bias, fairness, inclusivity, robustness, safety, and v...
Task 4.1 names Guardrails for Amazon Bedrock as the tool that identifies and steers responsible-AI features at inference. Lesson 4.1 taught the features. This lesson is the product. A guardrail is a versioned set of p...
Task 4.1 asks you to select a model with responsible practices. The exam examples are environmental considerations and sustainability. Domain 3 taught you to pick a catalog row for cost, modality, latency, and window....
Task 4.1 asks you to identify legal risks of working with generative AI. The exam list is intellectual property infringement claims, biased model outputs, loss of customer trust, end user risk, and hallucinations. Dom...
Task 4.1 asks you to identify characteristics of datasets. The exam list is inclusivity, diversity, curated data sources, and balanced datasets. A responsible model starts with the table, the clip pile, or the documen...
Task 4.1 asks you to understand the effects of bias and variance. The exam list is effects on demographic groups, inaccuracy, overfitting, and underfitting. Amazon Machine Learning's model-fit page is the first-party...
Task 4.1 closes the detective list. You already know the features, the dataset, and the fit failure. This lesson is the AWS toolkit that detects and monitors bias, trustworthiness, and truthfulness. The exam names ana...
Task 4.2 is transparent and explainable models. The first objective is the difference between a model you can explain and a model you cannot. AWS splits the words. Explainability is a mechanism to understand system be...
Task 4.2 names the tools that identify transparent and explainable models. The exam list is Amazon SageMaker Model Cards, open source models, data, and licensing. Lesson 4.8 taught the two words. This lesson is the pa...
Task 4.2 closes Domain 4 with two judgments. You identify tradeoffs between model safety and transparency. You understand principles of human-centered design for explainable AI. A stronger filter hides more text. A fu...
Domain 5 is Security, Compliance, and Governance for AI Solutions. It is 14 percent of scored AIF-C01 content. Task 5.1 starts with methods to secure AI systems. The official list names the AWS shared responsibility m...
Task 5.1 names IAM roles, policies, and permissions as the first AWS feature that secures an AI system. Lesson 5.1 placed identity on your side of the shared-responsibility line. This lesson is the product. IAM decide...
Task 5.1 names encryption among the AWS features that secure AI systems. The same task later lists encryption at rest and in transit as a privacy consideration. Lesson 5.1 put encryption on your side of the shared-res...
Task 5.1 names Amazon Macie and AWS PrivateLink next to IAM and encryption. Macie finds sensitive data in Amazon S3. PrivateLink keeps Bedrock API calls inside your VPC. Lesson 5.3 locked the disk and the TLS session....
Task 5.1 asks you to describe source citation and documenting data origins. The official examples are data lineage, data cataloging, and Amazon SageMaker Model Cards. Domain 4 already used Model Cards as a transparenc...
Task 5.1 asks for best practices for secure data engineering. The official list is assessing data quality, implementing privacy-enhancing technologies, data access control, and data integrity. Lesson 5.5 documented or...
Task 5.1 lists security and privacy considerations for AI systems. The official examples include application security, threat detection, vulnerability management, infrastructure protection, prompt injection, data leak...
Task 5.1 ends with hallucination detection methods and grounding techniques that improve output accuracy. The official examples are Retrieval Augmented Generation (RAG) grounding, output validation, and confidence sco...
Task 5.2 asks you to recognize governance and compliance regulations for AI systems. The older official exam guide names three families you must be able to spot: International Organization for Standardization (ISO), S...
Task 5.2 still has three jobs after the ISO and SOC papers. Identify the AWS services that help with governance and regulation compliance. Describe data governance strategies. Describe the processes that follow a gove...
Enroll in AWS AI Practitioner Study Notes today and get instant access to all 49 lessons, hands-on projects, and expert support.
Enroll Now - It's Free!