LevelUpWith

Explore / AI & Agents

AWS Certified AI Practitioner (AIF-C01) Exam Prep

Work through all five AIF-C01 domains and every task statement so you can pick the right AI, ML or generative AI approach and the right AWS service for a business scenario, and recognise when AI is the wrong tool. This independent course is not affiliated with AWS and follows exam guide version 1.1.

Read lesson 1 free →Try 10 practice questions free →

Levels 1 and 2 are free to read, no account needed. Levels 3 and 4 are free with a free account, and so are 20 answers from the AI tutor. 10 questions from the bank behind this course's timed mock are free to try, no account needed.

Three-dimensional cover artwork for “AWS Certified AI Practitioner (AIF-C01) Exam Prep”

Capable · 18 levels · 4 free · Created Oct 2026 · Professionally curated by levelupwith.com

What's inside

  1. Level 1: Exam Blueprint and Core AI TerminologyFree
    [Fundamentals of AI and ML] Sets out the exam facts for this independent, no-pass-promised course (exam guide version 1.1: 65 questions with 50 scored and 15 unscored, 90 minutes, scaled passing score 700, four question types, five weighted domains, out-of-scope tasks) and then defines the basic terms of task statement 1.1: AI, ML, deep learning, neural networks, computer vision, NLP, models, algorithms, training, inferencing, LLMs, GenAI and agentic AI.
    • State the exam format, scoring, question types, domain weightings and out-of-scope job tasks from exam guide version 1.1, and explain that this course is independent of AWS
    • Distinguish AI, ML, deep learning, GenAI and agentic AI and place each as a subset or extension of the others
    • Match core terms (model, algorithm, training, inferencing, bias, fairness, fit, LLM) to their definitions in a matching-style exercise
  2. Level 2: Data Types, Learning Types and InferencingFree
    [Fundamentals of AI and ML] Completes task statement 1.1 by classifying the data that models use (labeled and unlabeled, tabular, time-series, image, text, structured and unstructured), contrasting supervised, unsupervised and reinforcement learning, and distinguishing batch from real-time inferencing.
    • Classify a described dataset as labeled or unlabeled and as structured, unstructured, tabular, time-series, image or text
    • Choose supervised, unsupervised or reinforcement learning for a given scenario and explain why the other two do not fit
    • Decide between batch and real-time inferencing from a scenario's latency and volume requirements
    • Distinguish asynchronous and serverless inferencing from batch and real-time inferencing and choose the right type for a scenario's payload size, latency and traffic pattern.
  3. Level 3: Right Tool or Wrong Tool: Use Cases and AI Services
    [Fundamentals of AI and ML] Covers task statement 1.2: where AI/ML adds value, when it is the wrong tool (for example when a specific deterministic outcome is needed or the cost outweighs the benefit), when a traditional ML technique such as regression, classification or clustering beats a foundation model, and what each managed AI service is for (Amazon Comprehend, Lex, Polly, Rekognition, Textract, Transcribe, Translate, Personalize and SageMaker AI).
    • Recognise scenarios where AI/ML is not appropriate and a rule-based or deterministic solution is the better answer
    • Select regression, classification or clustering for a business problem, and say when a traditional model rather than an FM is the better fit
    • Match each managed AWS AI service to the business task it solves in product-contrast questions
    • Identify real-world AI applications such as fraud detection, forecasting, speech recognition, recommendation systems, knowledge bases and agentic AI in a business scenario.
  4. Level 4: The ML Lifecycle, MLOps and Model Metrics
    [Fundamentals of AI and ML] Covers task statement 1.3 at recognition level, with no pipeline building: the stages of the ML pipeline from data collection to monitoring, sources of models (open source pre-trained versus custom training), managed API versus self-hosted deployment, the SageMaker AI capabilities that serve each stage, MLOps concepts, and model versus business metrics.
    • Place the ML pipeline stages in the correct order in an ordering-style exercise and name what happens at each
    • Explain MLOps concepts such as repeatability, scalability, production readiness, monitoring and re-training, and identify which SageMaker AI capability supports a given stage
    • Tell model performance metrics (accuracy, AUC, F1 score) apart from business metrics (cost per user, ROI, customer feedback) and choose the relevant one for a scenario
    • Explain what precision and recall measure and choose between accuracy, precision, recall and F1 score for a scenario without calculating them.
  5. Level 5: GenAI Building Blocks: Tokens to Foundation Models
    [Fundamentals of GenAI] Covers the core concepts of task statement 2.1: tokens, chunking, embeddings and vectors, transformer-based LLMs, foundation models, multi-modal and diffusion models, typical GenAI use cases, and the foundation model lifecycle from data and model selection through pre-training, fine-tuning, evaluation, deployment and feedback.
    • Define tokens, chunking, embeddings and vectors and explain how each is used when a foundation model processes text
    • Distinguish transformer-based LLMs, multi-modal models and diffusion models and match each to a use case such as summarisation, chatbots, code or image generation
    • Put the foundation model lifecycle stages in order and describe the purpose of each
  6. Level 6: Context Engineering, Agentic AI and MCP
    [Fundamentals of GenAI] Finishes task statement 2.1 with the concepts added in guide version 1.1: prompt engineering versus context engineering, the context window, what makes a system agentic (reasoning, planning, tool use, memory, multi-step and multi-agent workflows), and the role of Model Context Protocol (MCP) in connecting models to tools and data.
    • Explain what context engineering is and how the context window limits what a model can use in one request
    • Describe the components of an AI agent and tell an agentic workflow apart from a single prompt-and-response interaction
    • State what Model Context Protocol (MCP) standardises and recognise a scenario where it is the relevant concept
  7. Level 7: GenAI Strengths, Limits and Business Value
    [Fundamentals of GenAI] Covers task statement 2.2: the advantages of GenAI (adaptability, responsiveness, simplicity), its disadvantages (hallucinations, interpretability, inaccuracy, nondeterminism), the factors for choosing a suitable model (model type, performance requirements, capabilities, constraints, compliance), and the business value metrics used to judge a GenAI application.
    • Identify the GenAI advantage or limitation at play in a business scenario, including hallucination and nondeterminism
    • Weigh model type, performance, capability, constraint and compliance factors to judge whether a GenAI model suits a requirement
    • Pick the business value metric (such as conversion rate, average revenue per user, efficiency, accuracy or customer lifetime value) that fits a stated business goal
    • Identify conversational capabilities and the ability to generate content as advantages of GenAI, alongside adaptability and responsiveness, in a business scenario.
  8. Level 8: AWS GenAI Services and Their Cost Tradeoffs
    [Fundamentals of GenAI] Covers task statement 2.3: what each AWS GenAI building block is for (Amazon Bedrock, Amazon Nova, Amazon SageMaker AI and SageMaker JumpStart, Amazon Quick, Kiro, Strands Agents, Amazon Bedrock AgentCore, AWS Transform), the compute and container services workloads can run on (Amazon EC2, AWS Lambda, Amazon ECS, Amazon EKS), the benefits of AWS infrastructure, and cost tradeoffs such as token-based pricing, provisioned throughput, responsiveness, availability and Regional coverage.
    • Match each in-scope GenAI service and developer tool, including the newest additions (Amazon Bedrock AgentCore, Amazon Nova, Kiro, Strands Agents, AWS Transform), to what it is for
    • Explain the advantages of building GenAI applications on AWS services, such as accessibility, lower barrier to entry, efficiency, cost-effectiveness and speed to market
    • Compare token-based pricing with provisioned throughput and custom-model costs, and choose the cost-appropriate option for a scenario
  9. Level 9: Selecting an FM, Inference Parameters and Agents
    [Applications of Foundation Models] Opens task statement 3.1 with the design choices for an FM-powered application: selection criteria for pre-trained models (cost, modality, latency, multi-lingual support, model size and complexity, customization, input and output length), the effect of inference parameters such as temperature and response length, and the role of agents in multi-step tasks, including Amazon Bedrock Agents and their business uses.
    • Choose a pre-trained foundation model for a scenario by weighing cost, modality, latency, language support and input/output length
    • Predict how changing temperature or maximum output length alters a model's responses and cost
    • Recognise when a multi-step business task calls for an agent and name the AWS agent capabilities (Amazon Bedrock Agents, Amazon Bedrock AgentCore, Strands Agents) that apply
    • Explain what prompt caching is and how it affects cost and latency when choosing a foundation model for an application.
  10. Level 10: RAG, Knowledge Bases and Vector Stores
    [Applications of Foundation Models] Continues task statement 3.1 with Retrieval Augmented Generation: how RAG grounds responses in company data, how Amazon Bedrock Knowledge Bases implements it, which AWS services can store embeddings (Amazon OpenSearch Service, Amazon Aurora, Amazon Neptune, Amazon RDS for PostgreSQL, Amazon DocumentDB), and how they differ from the other in-scope databases (Amazon DynamoDB, Amazon ElastiCache).
    • Describe the RAG flow in order — chunk, embed, store, retrieve, augment, generate — and state the business problems it solves
    • Explain what Amazon Bedrock Knowledge Bases manages on the builder's behalf
    • Select a suitable vector store for a scenario and say what each in-scope database service is primarily for
  11. Level 11: Customizing Foundation Models: Cost and Data
    [Applications of Foundation Models] Completes task statement 3.1 and covers task statement 3.3 at concept level: the cost and effort ladder from in-context learning through RAG, fine-tuning and model distillation to pre-training, the methods of fine-tuning (instruction tuning, domain adaptation, transfer learning, continuous pre-training), and how to prepare fine-tuning data (curation, governance, size, labeling, representativeness).
    • Rank in-context learning, RAG, fine-tuning, model distillation and pre-training by cost and effort, and choose the cheapest approach that meets a requirement
    • Distinguish instruction tuning, domain adaptation, transfer learning and continuous pre-training, and explain what model distillation achieves
    • Identify what makes a dataset fit for fine-tuning: curated, governed, sufficiently large, correctly labeled and representative
    • Describe reinforcement learning from human feedback (RLHF) and recognise when it is used to align a fine-tuned foundation model with human preferences.
  12. Level 12: Prompt Engineering Techniques and Risks
    [Applications of Foundation Models] Covers task statement 3.2: the parts of a prompt (instruction, context, negative prompts), techniques (zero-shot, single-shot, few-shot, chain-of-thought, prompt templates), best practices such as specificity, experimentation and guardrails, managing and versioning prompts with Amazon Bedrock Prompt Management, and the risks of prompt engineering (exposure, poisoning, hijacking, jailbreaking).
    • Identify the prompting technique used in an example prompt and choose the best technique for a given task
    • Apply prompt best practices and explain what Amazon Bedrock Prompt Management adds for reuse and versioning
    • Distinguish prompt exposure, poisoning, hijacking and jailbreaking in scenario questions
  13. Level 13: Evaluating Foundation Model Performance
    [Applications of Foundation Models] Covers task statement 3.4: evaluation approaches (human evaluation, benchmark datasets, Amazon Bedrock Model Evaluation, LLM-as-a-judge), the metrics ROUGE, BLEU and BERTScore and the tasks each suits, and how to judge whether an FM, RAG application or agent meets business objectives such as productivity, user engagement and task completion.
    • Choose between human evaluation, benchmark datasets, automatic evaluation and LLM-as-a-judge for a stated evaluation need
    • Match ROUGE, BLEU and BERTScore to summarisation, translation and semantic-similarity tasks without calculating them
    • Decide whether an FM application meets its business objectives using business alignment metrics rather than model metrics alone
  14. Level 14: Building Responsible AI Systems
    [Guidelines for Responsible AI] Covers task statement 4.1: the features of responsible AI (bias, fairness, inclusivity, robustness, safety, veracity), Amazon Bedrock Guardrails as a responsible-AI tool, responsible and sustainable model selection, legal risks of GenAI (intellectual property infringement, biased outputs, loss of customer trust, end-user risk, hallucinations), dataset characteristics (inclusive, diverse, balanced, curated), the effects of bias and variance (overfitting, underfitting, harm to demographic groups), and tools to detect and monitor bias such as Amazon SageMaker Clarify, Amazon SageMaker Model Monitor and human review with Amazon Augmented AI.
    • Name the features of responsible AI and identify which one a scenario violates
    • Recognise the legal risks of GenAI and the dataset characteristics that reduce bias, and explain overfitting and underfitting as effects of variance and bias
    • Select the tool that fits a responsible-AI need: Amazon Bedrock Guardrails, bias detection, model monitoring or human review
    • Describe how analyzing label quality, human audits and subgroup analysis detect and monitor bias, trustworthiness and truthfulness in a model.
  15. Level 15: Transparent and Explainable Models
    [Guidelines for Responsible AI] Covers task statement 4.2: the difference between transparent, explainable models and opaque ones, tools that document and assess models (Amazon SageMaker Model Cards, Amazon Bedrock Model Evaluations, open source models, data and licensing), the tradeoffs between model safety and transparency such as interpretability versus performance, and the principles of human-centered design for explainable AI.
    • Distinguish transparent and explainable models from opaque models and say when explainability is a requirement
    • Identify what Amazon SageMaker Model Cards and Amazon Bedrock Model Evaluations each contribute to transparency
    • Explain the tradeoff between interpretability and performance and apply human-centered design principles to an explainable-AI scenario
  16. Level 16: Securing AI Systems and Their Data
    [Security, Compliance, and Governance for AI Solutions] Covers task statement 5.1 at recognition level, not implementation: the shared responsibility model for AI, IAM roles, policies and permissions, encryption with AWS KMS, Amazon Macie for sensitive data discovery, AWS Secrets Manager, private connectivity with Amazon VPC and AWS PrivateLink, Amazon CloudFront, identity and policy controls for agents in Amazon Bedrock AgentCore, source citation and data lineage, secure data engineering with the in-scope analytics and storage services (AWS Glue, AWS Glue DataBrew, AWS Lake Formation, AWS Data Exchange, Amazon EMR, Amazon Redshift, Amazon S3, Amazon S3 Glacier), and threats such as prompt injection.
    • Match each security service or feature (IAM, AWS KMS, Amazon Macie, AWS Secrets Manager, Amazon VPC with AWS PrivateLink, Amazon Bedrock Guardrails) to the AI security need it addresses
    • Explain source citation, data lineage and data cataloging, and state what each in-scope analytics and storage service is for in secure data engineering
    • Recognise prompt injection and related threats in a scenario and choose the mitigating control, including grounding to detect hallucinations
    • Describe secure data engineering best practices (assessing data quality, privacy-enhancing technologies, data access control and data integrity) and recognise which one a scenario calls for.
    • Recognise the security and privacy considerations for AI systems, including data leakage prevention, output filtering and validation, audit logging of AI interactions, toxicity, threat detection and v
    • Describe how output validation and confidence scoring, alongside RAG grounding, detect hallucinations and improve output accuracy.
  17. Level 17: Governance and Compliance for AI Solutions
    [Security, Compliance, and Governance for AI Solutions] Covers task statement 5.2 at recognition level: regulatory compliance standards for AI, the services that support governance and compliance (AWS Config, Amazon Inspector, AWS Artifact, AWS CloudTrail, Amazon CloudWatch, AWS Trusted Advisor, AWS Well-Architected Tool), cost governance with AWS Budgets and AWS Cost Explorer, data governance strategies (lifecycle, logging, residency, monitoring, retention), and governance processes including the Generative AI Security Scoping Matrix.
    • Match each governance service to its purpose, for example AWS Artifact for compliance reports, AWS CloudTrail for API audit logs and AWS Config for configuration compliance
    • Describe data governance strategies such as data lifecycle, logging, residency, monitoring and retention, and choose AWS Budgets or AWS Cost Explorer for a cost-control need
    • Use the Generative AI Security Scoping Matrix to classify a GenAI workload and identify who is responsible for which controls
  18. Level 18: Timed Mock Exam: 65 Questions in 90 Minutes Timed mock
    A timed 65-question, 90-minute mock of original practice questions (not real exam questions) that tests all five AIF-C01 domains in proportion to their weightings: Fundamentals of AI and ML 20%, Fundamentals of GenAI 24%, Applications of Foundation Models 28%, Guidelines for Responsible AI 14%, and Security, Compliance, and Governance for AI Solutions 14%.
    Free 10 questions from the bank behind this course's timed mock are free to try, no account needed. Read all 10 with their answers →

Access

Levels 1 and 2 are free to read, no account needed. Levels 3 and 4 are free with a free account, and so are 20 answers from the AI tutor. 10 questions from the bank behind this course's timed mock are free to try, no account needed. Every level, the podcast edition and the AI tutor come with All Access at £4.99/month or any Creator plan — see pricing.

Read lesson 1 free →Try 10 practice questions free →

More like this