AIF-C01 Exam Domains Explained: What to Study First

Begin here even though this is not the largest domain. Later topics depend on understanding the relationship between artificial intelligence, machine learning, deep learning, and generative AI. Be able to distinguish supervised, unsupervised, and reinforcement learning, and recognise tasks such as classification, regression, and clustering. Connect each task to a business need: predicting a numerical value differs from assigning a category or grouping similar records. Study training and inference, labelled and unlabelled data, and batch versus real-time inference. The aim is not to memorise definitions in isolation, but to explain why a particular approach fits a particular problem.

Next, follow the machine learning lifecycle from defining the business objective through data preparation, training, evaluation, deployment, and monitoring. Understand overfitting, underfitting, and the purpose of separate training, validation, and test datasets. Recognise common evaluation measures, including accuracy, precision, recall, and F1 score, without turning revision into an advanced mathematics exercise. Learn the roles of relevant AWS services: Amazon SageMaker AI supports machine learning workflows, while services such as Amazon Comprehend, Amazon Textract, Amazon Transcribe, and Amazon Rekognition address specific capabilities. Also practise identifying situations where simpler rules or conventional analytics are more appropriate than machine learning.

Once the foundations are secure, study how generative AI differs from systems that primarily classify or predict. Understand foundation models, large language models, tokens, embeddings, and the basic purpose of transformer architectures at a conceptual level. Recognise that generative AI can produce text, images, and other content, and can support summarisation, conversational assistance, and content creation. A useful revision exercise is to take one business requirement and explain both why generative AI might help and why its output still needs evaluation. Plausible language is not proof of factual accuracy, and model capabilities depend on the task and deployment choices.

Give equal attention to capabilities and limitations. Study hallucinations, variability in responses, dependence on data quality, and the trade-offs between performance, latency, and cost. Recognise the lifecycle of a foundation model, including data selection, pre-training, adaptation, evaluation, deployment, and feedback. For AWS services, understand Amazon Bedrock as a managed service for building generative AI applications using foundation models, and distinguish that role from Amazon SageMaker AI and Amazon Q offerings. Focus on selecting the appropriate service category rather than memorising changing product catalogues. Recheck the official guide’s in-scope services before the exam because product names and capabilities can evolve.

Applications of Foundation Models is the largest domain, so it deserves the largest share of scenario-based practice after you understand the basics. Study model selection using requirements such as input and output modalities, response quality, latency, cost, and customisation needs. Then distinguish prompting, retrieval-augmented generation, and fine-tuning. Prompting shapes the request; retrieval-augmented generation supplies relevant retrieved information during the application workflow; fine-tuning adapts a model using additional training. These are not interchangeable solutions. Practise choosing between them when a company needs current internal knowledge, a particular response style, or improved performance on a specialised task with clearly defined evaluation criteria.

Build a practical understanding of prompt engineering, including clear instructions, relevant context, examples, and output constraints. Recognise zero-shot and few-shot prompting, and understand why poorly designed prompts can produce unreliable results. Study embeddings, vector databases, and how Amazon Bedrock Knowledge Bases supports retrieval workflows. Also recognise how agents can coordinate steps and use tools to fulfil requests. Finally, learn to evaluate model outputs against business objectives, using human assessment and appropriate metrics rather than assuming that fluent text is good text. Keep coding secondary: the priority is explaining architecture choices, their limitations, and their effect on the user’s requirements.

Do not treat responsible AI as optional reading because its weighting is lower. This domain asks you to recognise whether an AI solution is appropriate, fair, explainable, and sufficiently controlled for its intended use. Study bias, fairness, inclusivity, transparency, explainability, robustness, and safety. Understand that bias can originate in data collection, labelling, model development, or deployment conditions; it is not fixed simply by removing one sensitive field. Practise scenarios involving unrepresentative datasets, harmful content, and decisions that affect people. Your task is to identify the risk and choose a suitable mitigation, while recognising that no single control guarantees responsible behaviour.

Connect principles with concrete techniques and AWS capabilities. Representative data, documented limitations, human review, and ongoing evaluation all support responsible deployment. Understand the role of Amazon SageMaker Clarify in detecting potential bias and helping explain model predictions, and recognise Amazon Bedrock Guardrails as a way to apply configurable safeguards to generative AI interactions. Guardrails do not make every response correct or remove the need for testing. Learn the distinction between explaining how a system behaves and proving that its outcomes are fair. In revision, revisit earlier use cases and ask who could be harmed, what evidence is needed, and where human oversight belongs.

Security, Compliance, and Governance for AI Solutions covers protecting AI systems and managing how they are used. Study the AWS shared responsibility model, least-privilege access with AWS Identity and Access Management, encryption, and data protection. Understand why sensitive information requires controls across inputs, stored data, retrieved content, and outputs. Recognise prompt injection as a security concern, including malicious instructions embedded in material an application retrieves. Connect relevant AWS services to their roles: AWS CloudTrail records account activity, Amazon Macie helps discover sensitive data in Amazon S3, and AWS Artifact provides access to compliance reports and agreements. Service recognition should support reasoning, not replace it.

For governance, revise data provenance, documentation, auditability, monitoring, and policies governing model and data use. Using an AWS service does not automatically make a workload compliant; obligations depend on the organisation, configuration, data, and applicable requirements. Once you have covered all five AWS AI Practitioner exam domains, use their weightings as a starting point for distributing review time, then adjust for weaknesses. A 14% domain where you repeatedly misunderstand scenarios may need more attention than a stronger 28% domain. Use [Erudex practice tests](/practice) to identify gaps and explain incorrect choices. Any course-completion certificate should be distinguished from AWS Certification, which requires passing the AWS exam.