AWS Certified AI Practitioner (AIF-C01) Fact Sheet¶

Exam Overview¶

Exam Code: AIF-C01 Exam Name: AWS Certified AI Practitioner Duration: 90 minutes Questions: 65 questions Question Format: Multiple choice and multiple response Passing Score: 700/1000 (scaled scoring, approximately 70%) Cost: $150 USD (50% discount for previous AWS certification holders) Valid For: 3 years Prerequisites: None (foundational-level certification) Language: Available in English, with more languages coming Delivery: Pearson VUE (online proctored or testing center)

📖 Official Exam Page - Registration and details 📖 Exam Guide PDF - Detailed objectives 📖 Sample Questions - Official practice questions

Target Audience¶

This certification is designed for: - Business analysts seeking AI/ML knowledge for decision-making - Project managers leading AI/ML initiatives - Sales professionals selling AI solutions - Marketing professionals leveraging AI tools - Technical professionals starting their AI/ML journey - Students entering the AI field

No coding experience required - Focus is on concepts, business applications, and AWS AI services.

📖 AWS AI/ML Learning Path - Official learning path 📖 Getting Started with AI on AWS - AI overview

Exam Domains¶

Domain 1: Fundamentals of AI and ML (20%)¶

This foundational domain covers core AI/ML concepts and terminology.

1.1 Explain Basic AI Concepts¶

Key Concepts: - Artificial Intelligence (AI) vs Machine Learning (ML) vs Deep Learning - Common AI terminology and vocabulary - Types of AI problems (classification, regression, clustering) - AI/ML workflow and lifecycle

📖 What is Artificial Intelligence? - AI overview 📖 What is Machine Learning? - ML fundamentals 📖 What is Deep Learning? - Deep learning explained 📖 Machine Learning Pipeline - ML workflow

1.2 Types of Machine Learning¶

Supervised Learning: - Training with labeled data - Classification (discrete outputs) - Regression (continuous outputs) - Examples: Image classification, price prediction

📖 Supervised Learning - Supervised ML overview 📖 Amazon SageMaker Algorithms - Built-in algorithms

Unsupervised Learning: - Training with unlabeled data - Clustering (grouping similar data) - Dimensionality reduction - Anomaly detection

📖 Unsupervised Learning - Unsupervised methods 📖 K-Means Clustering - Clustering algorithm

Reinforcement Learning: - Learning through trial and error - Reward-based optimization - Agent-environment interaction - Applications in robotics, gaming

📖 Reinforcement Learning - RL on AWS 📖 AWS DeepRacer - RL learning platform

1.3 ML Development Lifecycle¶

Key Stages: 1. Problem Definition: Identify business objectives 2. Data Collection: Gather relevant data 3. Data Preparation: Clean and transform data 4. Model Training: Train ML models 5. Model Evaluation: Test model performance 6. Deployment: Deploy to production 7. Monitoring: Track model performance over time

📖 ML Workflow - Complete ML lifecycle 📖 Amazon SageMaker - End-to-end ML platform 📖 SageMaker Studio - Integrated development environment 📖 SageMaker Pipelines - ML workflows

1.4 Model Evaluation and Metrics¶

Key Concepts: - Training vs Validation vs Test datasets - Overfitting vs Underfitting - Bias-Variance tradeoff - Common metrics: Accuracy, Precision, Recall, F1-Score, AUC-ROC

📖 Model Evaluation - Evaluation concepts 📖 SageMaker Model Monitor - Monitoring models 📖 SageMaker Clarify - Model explainability

Domain 2: Fundamentals of Generative AI (24%)¶

The largest domain, covering generative AI concepts and foundation models.

2.1 Generative AI Concepts¶

What is Generative AI: - Creates new content (text, images, audio, video) - Foundation models and large language models - Transformer architecture basics - Differences from traditional ML

📖 What is Generative AI? - GenAI overview 📖 Generative AI on AWS - AWS GenAI services 📖 Foundation Models Explained - Foundation model concepts

Types of Generative AI Models: - Large Language Models (LLMs): Text generation and understanding - Diffusion Models: Image and video generation - Generative Adversarial Networks (GANs): Synthetic data generation - Variational Autoencoders (VAEs): Data encoding and generation

📖 Large Language Models - LLM overview 📖 Stable Diffusion on AWS - Image generation

2.2 Prompt Engineering¶

Prompt Engineering Techniques: - Zero-shot prompting - Few-shot prompting (providing examples) - Chain-of-thought prompting - Prompt templates and best practices - System prompts vs user prompts

📖 Prompt Engineering Guide - Bedrock prompting 📖 Prompt Engineering Best Practices - Prompting techniques 📖 Anthropic Claude Prompting - Claude-specific guidance

Prompt Optimization: - Clear and specific instructions - Context and constraints - Output format specification - Iterative refinement - Temperature and sampling parameters

📖 Inference Parameters - Model configuration 📖 Text Generation Parameters - Parameter tuning

2.3 Retrieval Augmented Generation (RAG)¶

RAG Architecture: - Combines retrieval with generation - Vector databases for semantic search - Embeddings and similarity matching - Reduces hallucinations with grounded data - Use cases: Knowledge bases, chatbots, Q&A systems

📖 What is RAG? - RAG explained 📖 Amazon Bedrock Knowledge Bases - Managed RAG solution 📖 Knowledge Base Setup - Creating knowledge bases 📖 Amazon Kendra - Intelligent search 📖 Amazon OpenSearch Service - Vector search

Embeddings and Vector Databases: - Text embeddings for semantic similarity - Vector storage and retrieval - Similarity search algorithms - AWS services for vector search

📖 Amazon Bedrock Text Embeddings - Embedding models 📖 Vector Engine for OpenSearch - K-NN search 📖 Amazon RDS with pgvector - PostgreSQL vector extension

2.4 Fine-Tuning and Customization¶

Model Customization Approaches: - Prompt Engineering: Modify inputs (no training) - In-Context Learning: Few-shot examples - Fine-Tuning: Train on domain-specific data - Custom Model Training: Build from scratch

📖 Model Customization - Bedrock customization 📖 Fine-Tuning Guide - Fine-tuning process 📖 Continued Pre-training - Pre-training customization

Domain 3: Applications of Foundation Models (28%)¶

The highest-weighted domain, covering Amazon Bedrock and model selection.

3.1 Amazon Bedrock Overview¶

What is Amazon Bedrock: - Fully managed service for foundation models - Access to multiple model providers - Serverless and pay-per-use - Private and secure - No infrastructure management

📖 Amazon Bedrock - Service overview 📖 Bedrock Features - Key capabilities 📖 Bedrock Pricing - Cost structure 📖 Getting Started with Bedrock - Quick start

Bedrock Components: - Foundation models - Knowledge bases (RAG) - Agents (task automation) - Guardrails (safety controls) - Model evaluation

📖 Bedrock Agents - Autonomous agents 📖 Bedrock Guardrails - Content filtering 📖 Model Evaluation - Comparing models

3.2 Foundation Model Providers¶

Available Models in Bedrock:

Anthropic Claude: - Claude 3 (Haiku, Sonnet, Opus) - Long context windows (200K+ tokens) - Strong reasoning capabilities - Best for: Complex analysis, coding, detailed content

📖 Claude Models - Claude on Bedrock 📖 Claude 3 Model Family - Model comparison

Amazon Titan: - Text generation (Express, Lite) - Image generation - Embeddings (Text, Multimodal) - Cost-effective, native AWS models

📖 Amazon Titan Models - Titan family 📖 Titan Text - Text models 📖 Titan Embeddings - Embedding models 📖 Titan Image Generator - Image generation

AI21 Labs Jurassic: - Jurassic-2 models - Multilingual capabilities - Good for instruction following

📖 AI21 Jurassic - Jurassic models

Cohere: - Command models for chat and generation - Embed models for embeddings - Multilingual support

📖 Cohere Models - Cohere on Bedrock

Meta Llama: - Llama 2 and Llama 3 models - Open-source foundation - Various sizes (7B, 13B, 70B parameters)

📖 Meta Llama - Llama models

Stability AI: - Stable Diffusion XL - Image generation - SDXL 1.0 model

📖 Stability AI Models - Stable Diffusion

Mistral AI: - Mistral and Mixtral models - Efficient and performant - Open-source based

📖 Mistral Models - Mistral on Bedrock

3.3 Model Selection Criteria¶

Factors to Consider: - Use Case: Text, image, code generation - Context Length: Token limits for inputs - Latency: Response time requirements - Cost: Token pricing and throughput - Accuracy: Model performance on task - Language Support: Multilingual needs

📖 Supported Models and Regions - Model availability 📖 Model IDs - Identifying models 📖 Choosing a Model - Selection guide

3.4 Other AWS AI/ML Services¶

Amazon SageMaker: - Build, train, and deploy ML models - Jupyter notebooks and Studio - Pre-built algorithms - Model registry and MLOps

📖 Amazon SageMaker - ML platform 📖 SageMaker JumpStart - Pre-trained models 📖 SageMaker Canvas - No-code ML

Amazon Q (AI Assistant): - Amazon Q Developer (for coding) - Amazon Q Business (enterprise search) - Conversational AI powered by LLMs

📖 Amazon Q Developer - AI coding assistant 📖 Amazon Q Business - Enterprise AI assistant

Amazon CodeWhisperer: - AI code generator (now part of Amazon Q Developer) - Real-time code suggestions - Security scans

📖 Amazon CodeWhisperer - Code generation

AWS AI Services (Pre-Trained APIs): - Amazon Rekognition: Image and video analysis - Amazon Textract: Document text extraction - Amazon Comprehend: Natural language processing - Amazon Transcribe: Speech to text - Amazon Polly: Text to speech - Amazon Translate: Language translation - Amazon Lex: Conversational interfaces

📖 Amazon Rekognition - Computer vision 📖 Amazon Textract - Document analysis 📖 Amazon Comprehend - NLP service 📖 Amazon Transcribe - Speech recognition 📖 Amazon Polly - Text-to-speech 📖 Amazon Translate - Translation 📖 Amazon Lex - Chatbots

Domain 4: Guidelines for Responsible AI (14%)¶

Critical domain covering AI ethics, bias, and fairness.

4.1 AI Ethics and Responsible AI Principles¶

Core Principles: - Fairness: Avoid bias and discrimination - Transparency: Explain AI decisions - Privacy: Protect user data - Security: Prevent misuse - Accountability: Human oversight - Reliability: Consistent performance

📖 AWS Responsible AI - AWS principles 📖 Responsible Use of AI - Best practices 📖 AI Service Cards - Service transparency

4.2 Bias and Fairness¶

Types of Bias: - Data Bias: Biased training data - Algorithmic Bias: Model design issues - Human Bias: Labeling and interpretation - Deployment Bias: Unequal access or application

Bias Detection and Mitigation: - Diverse training datasets - Fairness metrics and evaluation - Regular audits and testing - Stakeholder involvement

📖 SageMaker Clarify - Bias detection 📖 Detecting Bias - Pre-training bias 📖 Post-Training Bias Metrics - Model fairness

4.3 Model Explainability and Transparency¶

Explainability Techniques: - Feature importance - SHAP (SHapley Additive exPlanations) - Model cards and documentation - Decision trees and interpretable models

📖 Model Explainability - SageMaker Clarify 📖 SHAP Explanations - Feature attribution

4.4 AI Governance¶

Governance Framework: - Clear policies and procedures - Role-based access control - Model versioning and tracking - Incident response plans - Continuous monitoring

📖 AI Governance - Governance best practices 📖 SageMaker Model Cards - Model documentation 📖 SageMaker Model Registry - Model versioning

4.5 Bedrock Guardrails¶

Content Filtering: - Denied topics - Content filters (hate, violence, sexual, etc.) - Word and phrase blocking - PII redaction - Hallucination detection

📖 Bedrock Guardrails - Safety controls 📖 Creating Guardrails - Configuration 📖 Content Filters - Filtering options

Domain 5: Security, Compliance, and Governance (14%)¶

Covers data protection, security, and regulatory compliance.

5.1 Data Security and Privacy¶

Data Protection: - Encryption at rest and in transit - Data isolation and privacy - Access controls - Data retention policies - Compliance certifications

📖 Bedrock Data Protection - Security overview 📖 Bedrock Encryption - Encryption details 📖 AWS KMS - Key management

Privacy Considerations: - No data used for model training (Bedrock promise) - Data residency and sovereignty - GDPR, CCPA compliance - PII handling and redaction

📖 Bedrock Privacy - Privacy commitments 📖 Compliance Programs - AWS compliance

5.2 Identity and Access Management¶

Access Control: - IAM policies for Bedrock - Role-based access control (RBAC) - Service control policies (SCPs) - Cross-account access

📖 Bedrock IAM - Access management 📖 Identity-Based Policies - Policy examples 📖 Service Roles - IAM roles

5.3 Monitoring and Logging¶

Observability: - CloudWatch metrics and logs - CloudTrail API logging - Model invocation logging - Cost tracking

📖 Bedrock Monitoring - Monitoring overview 📖 CloudWatch Metrics - Available metrics 📖 CloudTrail Logging - API logging 📖 Invocation Logging - Request/response logs

5.4 Compliance and Regulatory Requirements¶

Compliance Frameworks: - SOC 1, 2, 3 - ISO 27001, 27017, 27018 - PCI DSS - HIPAA (for eligible services) - FedRAMP (coming soon)

📖 AWS Compliance - Compliance overview 📖 Bedrock Compliance - Service compliance

Study Strategy¶

Recommended Timeline (4-6 weeks, 8-12 hours/week)¶

Weeks 1-2: AI/ML Fundamentals - Study AI/ML basic concepts - Learn types of machine learning - Understand ML workflow - Review AWS AI service overview - Study time: 8-10 hours/week

Weeks 3-4: Generative AI and Bedrock - Deep dive into generative AI concepts - Hands-on with Amazon Bedrock - Practice prompt engineering - Learn about RAG and embeddings - Study time: 10-12 hours/week

Weeks 5-6: Responsible AI and Review - Study AI ethics and bias - Learn security and compliance - Take practice exams (aim for 75%+) - Review weak areas - Study time: 8-10 hours/week

Study Resources¶

Official AWS Training: 📖 AWS Skill Builder - Free AWS training 📖 AI Practitioner Learning Plan - Official study plan 📖 Generative AI Learning Plan - GenAI fundamentals

Hands-On Practice: 📖 Bedrock Playground - Try Bedrock models 📖 AWS PartyRock - No-code AI app builder 📖 Bedrock Samples on GitHub - Code examples

Additional Resources: - AWS AI/ML Blog posts - AWS re:Invent videos on AI/ML - Generative AI documentation - Practice with ChatGPT, Claude, or other LLMs

Exam Day Tips¶

Preparation¶

• Review key AI/ML terminology
• Understand Bedrock model providers and use cases
• Know responsible AI principles
• Get good sleep before exam
• Arrive/start 15-30 minutes early

During Exam¶

• Read questions carefully - focus on business context
• Look for keywords: "MOST appropriate", "BEST practice", "LEAST cost"
• Eliminate obviously wrong answers
• Flag uncertain questions for review
• Manage time: ~1.4 minutes per question
• Trust your preparation

Common Question Patterns¶

• Business use case selection (which AI service?)
• Model selection for specific tasks
• Responsible AI scenarios (bias, ethics, transparency)
• RAG vs fine-tuning vs prompt engineering
• Security and compliance requirements

Technical Setup (Online Proctoring)¶

• Stable internet connection
• Webcam and microphone required
• Clear workspace
• Government-issued photo ID
• Close all other applications

📖 Exam Prep Tips - Official guidance

After Certification¶

Career Benefits¶

• Demonstrates AI/ML knowledge to employers
• Foundation for technical AI certifications
• Validates understanding of AWS AI services
• Opens doors to AI-focused roles

Next Certifications¶

📖 AWS Certified Machine Learning - Specialty - Technical ML certification 📖 AWS Certified Data Analytics - Specialty - Data and ML pipeline focus 📖 AWS Certified Solutions Architect - Associate - Architecture foundation

Continuous Learning¶

• Follow AWS AI/ML blog
• Experiment with new Bedrock models
• Join AWS AI/ML communities
• Attend re:Invent and AI/ML sessions
• Build AI projects

📖 AWS Machine Learning Blog - Latest updates 📖 AWS AI/ML Newsletter - Monthly updates

Quick Reference¶

Exam Details at a Glance¶

• 65 questions in 90 minutes = ~1.4 minutes per question
• 700/1000 to pass = Approximately 70%
• 20% AI/ML fundamentals = ~13 questions
• 24% GenAI fundamentals = ~16 questions
• 28% Foundation model applications = ~18 questions
• 14% Responsible AI = ~9 questions
• 14% Security & compliance = ~9 questions

Key AWS Services to Know¶

Bedrock Model Providers Quick Reference¶

Good luck with your AWS Certified AI Practitioner exam! 🎉🤖