AIF-C01: Quick Refresher

← Back to Overview

Final Review

This page is designed for the final "cram" session before stepping into the AIF-C01 exam.

🏗️ Domain 1: AI/ML Fundamentals (20%)

AI vs. ML vs. DL

AI: Broadest category (mimicking human intelligence).
ML: Learning from data without explicit rules.
Deep Learning (DL): Multi-layered neural networks (mimics brain).

Learning Types

Supervised: Uses labeled data (e.g., house price prediction).
Unsupervised: Uses unlabeled data (e.g., customer clustering).
Reinforcement: Learns via rewards/penalties (e.g., AWS DeepRacer).

The ML Pipeline

Business Goal ➡️ 2. Data Prep ➡️ 3. Model Training ➡️ 4. Evaluation ➡️ 5. Deployment (Inference) ➡️ 6. Monitoring.

Inference Types

Real-time: Low latency, immediate (e.g., fraud check at checkout).
Batch: High volume, delayed (e.g., generating monthly recommendations).

✨ Domain 2 & 3: Generative AI & Foundation Models (52%)

Foundation Models (FMs)

Huge models pre-trained on massive data; multi-purpose.

Key GenAI Concepts

Tokens: Units of text (not always full words).
Hallucination: Confidently wrong output.
Temperature: Creativity setting (0 = predictable, 1 = creative).

Model Customization

Prompt Engineering: Designing better inputs (Zero-shot, Few-shot).
RAG (Retrieval-Augmented Generation): Connecting the model to external data (Vector DB) for up-to-date, private info.
Fine-tuning: Re-training the model on specific data to change its weight/behavior.

Evaluation Metrics

ROUGE/BLEU: Measures text similarity (mostly for summarization/translation).

🛠️ AWS Services Comparison (The "Which Tool" Section)

Service	Primary Use Case
Amazon Bedrock	API-based access to Foundation Models (Claude, Llama, Titan). Fastest for GenAI.
Amazon SageMaker	The "Kitchen Sink." Full control over building, training, and deploying custom models.
Amazon Q	AI-powered assistant for businesses (Q Business) or developers (Q Developer).
Rekognition	Computer Vision (image/video analysis).
Polly / Transcribe	Text-to-Speech (Polly) / Speech-to-Text (Transcribe).
Comprehend	Natural Language Processing (sentiment analysis, entity extraction).
Lex	Building conversational bots (chatbots).

🛡️ Domain 4 & 5: Security & Responsible AI (28%)

Responsible AI Pillars

Fairness, Explainability, Privacy, Robustness, Governance.

AWS Tools for Responsibility

SageMaker Clarify: Detects bias and provides model explainability.
Bedrock Guardrails: Filters out harmful content or PII from LLM responses.
Amazon A2I: Adds a "Human-in-the-loop" for reviewing low-confidence predictions.

Security & Compliance

Shared Responsibility: AWS secures the "of" (hardware), You secure the "in" (data/settings).
Data Privacy: Bedrock does not use customer data to train its base models.
Governance: Model Cards (documentation) and AI Service Cards provide transparency.

🎯 Bedrock vs SageMaker Quick Decision Tree

Need to use AI/ML?
├─ Want pre-built models via API?
│  ├─ GenAI/LLMs? → Amazon Bedrock ⭐
│  └─ Specific tasks? → AI Services (Rekognition, Comprehend, etc.)
│
└─ Want to build/train custom models?
   └─ Full ML lifecycle control? → SageMaker ⭐

When to Choose Bedrock

✅ Need LLMs quickly (no training)
✅ Text/Chat generation
✅ API-first approach
✅ Multi-model access (Claude, Llama, Titan)
✅ RAG implementation
✅ Quick prototyping

When to Choose SageMaker

✅ Custom model training
✅ Full control over ML pipeline
✅ Data science workflows
✅ Model monitoring & drift detection
✅ Specialized use cases
✅ MLOps requirements

🔑 Key Acronyms to Know

Acronym	Full Form	Quick Definition
FM	Foundation Model	Large pre-trained model
LLM	Large Language Model	Text-focused foundation model
RAG	Retrieval-Augmented Generation	Connect LLM to external data
A2I	Amazon Augmented AI	Human review workflows
MLOps	ML Operations	DevOps for ML models
PII	Personally Identifiable Information	Sensitive personal data
ROUGE	Recall-Oriented Understudy for Gisting Evaluation	Text similarity metric

💡 Final Minute Tips

Service Selection Rules

If the question asks for "Easy/No-Code/API": Think Bedrock or High-level AI services (Rekognition, Polly).
If the question asks for "Full Control/Data Scientist": Think SageMaker.
If the question asks about "Bias": Think SageMaker Clarify.
If the question asks about "External Knowledge/Real-time data": Think RAG or Knowledge Bases for Bedrock.

Common Exam Traps

Watch Out!

Bedrock ≠ Training: Bedrock uses pre-trained models only
Hallucinations: LLMs can be confidently wrong - use RAG or guardrails
Temperature: Higher = creative but less accurate
Fine-tuning ≠ Prompt Engineering: Fine-tuning changes the model; prompting doesn't
Shared Responsibility: You're responsible for data, AWS handles infrastructure

📊 Model Selection Quick Guide

For Text Tasks

Task	Best Service	Why
Chat/Conversation	Bedrock (Claude)	Natural dialogue
Code Generation	Bedrock (Claude/CodeWhisperer)	Optimized for code
Summarization	Bedrock (Titan/Claude)	Fast, accurate
Translation	Translate (simple) / Bedrock (complex)	Cost vs capability
Sentiment	Comprehend	Purpose-built

For Vision Tasks

Task	Best Service	Why
Object Detection	Rekognition	Pre-built, easy
Face Analysis	Rekognition	Specialized
Custom Vision	SageMaker	Full control
Medical Imaging	SageMaker	Compliance needs

🎓 Responsible AI Quick Checks

Before Deployment Checklist

[ ] Fairness: Does model work equally for all groups?
[ ] Explainability: Can you explain decisions?
[ ] Privacy: Is sensitive data protected?
[ ] Safety: Are guardrails in place?
[ ] Transparency: Is model documented?
[ ] Monitoring: Is drift detection enabled?

Key Questions to Answer

Is there bias? → Use SageMaker ClarifyNeed human review? → Use Amazon A2IFilter harmful output? → Use Bedrock GuardrailsTrack model performance? → Use SageMaker Model Monitor

🚀 RAG Implementation Quick Reference

Components

User Query
    ↓
1. Embedding Model (convert query to vector)
    ↓
2. Vector Database (find similar documents)
    ↓
3. Retrieved Context + Query
    ↓
4. LLM (generate answer with context)
    ↓
Answer with Sources

AWS RAG Stack

Vector DB: OpenSearch, Bedrock Knowledge Bases
Embeddings: Bedrock Titan Embeddings
LLM: Bedrock (Claude, Llama)
Storage: S3 (documents)

🎬 Additional Resources

Essential Video

Watch this AWS Certified AI Practitioner Exam Guide

This video provides a direct comparison between the two heavy-hitters of the exam—Bedrock and SageMaker—helping you decide which service fits specific exam scenarios.

⚡ Last 5 Minutes Before Exam

Must Remember

Bedrock = API access to FMs (no training)
SageMaker = Full ML lifecycle (build/train/deploy)
RAG = External knowledge for LLMs
Clarify = Bias detection and explainability
Shared Responsibility = AWS hardware, You data

Quick Mental Check

Can you explain AI vs ML vs DL? ✓
Do you know when to use Bedrock vs SageMaker? ✓
Can you describe RAG in one sentence? ✓
Do you know the 6 Responsible AI principles? ✓
Can you name 3 AI services besides Bedrock/SageMaker? ✓

You've Got This!

Take a deep breath. You've studied. Trust your preparation. Good luck! 🍀

← Back to Overview | Study Notes | Exam Tips

Last Updated: 2026-01-14

AIF-C01: Quick Refresher ​

🏗️ Domain 1: AI/ML Fundamentals (20%) ​

AI vs. ML vs. DL ​

Learning Types ​

The ML Pipeline ​

Inference Types ​

✨ Domain 2 & 3: Generative AI & Foundation Models (52%) ​

Foundation Models (FMs) ​

Key GenAI Concepts ​

Model Customization ​

Evaluation Metrics ​

🛠️ AWS Services Comparison (The "Which Tool" Section) ​

🛡️ Domain 4 & 5: Security & Responsible AI (28%) ​

Responsible AI Pillars ​

AWS Tools for Responsibility ​

Security & Compliance ​

🎯 Bedrock vs SageMaker Quick Decision Tree ​

When to Choose Bedrock ​

When to Choose SageMaker ​

🔑 Key Acronyms to Know ​

💡 Final Minute Tips ​

Service Selection Rules ​

Common Exam Traps ​

📊 Model Selection Quick Guide ​

For Text Tasks ​

For Vision Tasks ​

🎓 Responsible AI Quick Checks ​

Before Deployment Checklist ​

Key Questions to Answer ​

🚀 RAG Implementation Quick Reference ​

Components ​

AWS RAG Stack ​

🎬 Additional Resources ​

Essential Video ​

⚡ Last 5 Minutes Before Exam ​

Must Remember ​

Quick Mental Check ​