AWS Certified AI Practitioner (AIF-C01) — 48 Hour Study Guide

The Ultimate Crash Course: Pass the AWS AI Practitioner Exam in 48 Hours Using Battle-Tested Mnemonics

---

📋 Exam Overview

• Duration: 90 minutes
• Questions: 65 total (50 scored + 15 unscored)
• Passing Score: 700/10

---

📊 Domain Breakdown & Weightings

Domain	Weight	Hours to Allocate
Domain 1: Fundamentals of AI and ML	20%	10 hours
Domain 2: Fundamentals of Generative AI	24%	12 hours
Domain 3: Applications of Foundation Models	28%	14 hours
Domain 4: Guidelines for Responsible AI	14%	7 hours
Domain 5: Security, Compliance, and Governance	14%	5 hours

---

🗓️ 48-Hour Study Schedule

Day 1 (24 Hours)

• Hours 1–10: Domain 1 — Fundamentals of AI and ML
• Hours 11–22: Domain 2 — Fundamentals of Generative AI
• Hours 23–24: Break & Review

Day 2 (24 Hours)

• Hours 25–38: Domain 3 — Applications of Foundation Models
• Hours 39–45: Domain 4 — Guidelines for Responsible AI
• Hours 46–48: Domain 5 — Security, Compliance, and Governance

---

🧠 MEGA MNEMONICS & MEMORY TRICKS

Domain 1: AI/ML Fundamentals — "SMART CATS"

1. Supervised (labeled data)
2. Machine Learning (subset of AI)
3. Algorithms (instructions for models)
4. Reinforcement (reward-based learning)
5. Training (learning phase)
6. Computer Vision (image processing)
7. Analysis (EDA — Exploratory Data Analysis)
8. Time-series (sequential data)
9. Structured data (organized format)

Domain 2: Generative AI — "BEAM PROMPT"

1. Bedrock (AWS foundation model service)
2. Embeddings (vector representations)
3. Augmented (RAG — Retrieval Augmented Generation)
4. Multi-modal (text, image, audio)
5. Prompt (input instructions)
6. Response (model output)
7. Optimization (fine-tuning)
8. Models (foundation models)
9. Parameters (temperature, top-k)
10. Tokens (text units)

Domain 3: Foundation Models — "FINE RAGE"

1. Fine-tuning (customization)
2. Inference (prediction phase)
3. Neural networks (deep learning)
4. Evaluation (performance metrics)
5. RAG (Retrieval Augmented Generation)
6. Agents (multi-step tasks)
7. Guardrails (safety measures)
8. Embeddings (vector storage)

Domain 4: Responsible AI — "SAFE BET"

1. Safety (harm prevention)
2. Accountability (transparency)
3. Fairness (bias reduction)
4. Explainability (interpretable)
5. Bias (unfair treatment)
6. Ethics (moral considerations)
7. Trustworthy (reliable)

Domain 5: Security & Governance — "COMPLY IAM"

1. Compliance (regulations)
2. Organization (governance)
3. Monitoring (observability)
4. Policies (access control)
5. Lineage (data tracking)
6. Yield (audit results)
7. IAM (Identity & Access Management)
8. Artifact (compliance reports)
9. Macie (data discovery)

---

📚 CHEATSHEET: Key AWS Services by Domain

Domain 1: Core AI/ML Services — "SPLIT TECH"

1. SageMaker (end-to-end ML)
2. Polly (text-to-speech)
3. Lex (chatbots)
4. Inspector (vulnerability assessment)
5. Transcribe (speech-to-text)
6. Translate (language translation)
7. EC2 (compute instances)
8. Comprehend (text analysis)
9. Human-in-the-loop (A2I)

Domain 2: Generative AI Services — "JUMP BEDROCK"

1. JumpStart (pre-trained models)
2. User interface (PartyRock)
3. Models (foundation models)
4. Platform (SageMaker)
5. Bedrock (foundation model service)
6. Embeddings (vector storage)
7. Development (Q Developer)
8. RAG (retrieval augmented)
9. Optimization (fine-tuning)
10. Chatbots (conversational AI)
11. Knowledge bases (vector DBs)

Domain 3: Vector Databases & Storage — "OPEN ROADS"

1. OpenSearch (vector search)
2. PostgreSQL (pgvector extension)
3. Embeddings (vector storage)
4. Neptune (graph database)
5. RDS (relational database)
6. Optimization (performance)
7. Aurora (serverless DB)
8. DocumentDB (MongoDB compatible)
9. S3 (object storage)

---

🎯 KEY CONCEPTS CHEATSHEET

ML Types & Use Cases

Type	Data	Use Case	Example
Supervised	Labeled	Prediction	Email spam detection
Unsupervised	Unlabeled	Pattern discovery	Customer segmentation
Reinforcement	Reward-based	Decision making	Game playing AI

Generative AI Model Types

Model Type	Input	Output	AWS Service
LLM	Text	Text	Bedrock (Claude, Titan)
Diffusion	Text	Image	Bedrock (Stable Diffusion)
Multi-modal	Text + Image	Text/Image	Bedrock (Claude 3)

Inference Types

• Batch: Process large datasets offline (cost-effective)
• Real-time: Immediate response (higher cost, low latency)
• Streaming: Continuous processing (live data)

Prompt Engineering Techniques — "ZERO CHAIN FEW"

• ZERO-shot: No examples
• CHAIN-of-thought: Step-by-step reasoning
• FEW-shot: Multiple examples

---

🔧 AWS SERVICE DEEP DIVE

Amazon Bedrock (Critical!)

Foundation Models Available:

• Anthropic Claude (text generation, conversation)
• Amazon Titan (text, embeddings)
• Stability AI (image generation)
• AI21 Labs Jurassic (text generation)
• Cohere Command (text generation)
• Meta Llama (text generation)

Key Features:

• Serverless foundation models
• No infrastructure management
• Custom model fine-tuning
• Agents for complex tasks
• Knowledge bases with RAG
• Guardrails for safety

Amazon SageMaker Components — "JUMP FEATURE CLARIFY"

1. JUMPStart (pre-trained models)
2. Feature Store (feature management)
3. Endpoints (model hosting)
4. Autopilot (AutoML)
5. Training jobs (model training)
6. User interface (Studio)
7. Reprocessing (data prep)
8. Experiments (tracking)
9. Clarify (bias detection)
10. Lifecycle (MLOps)
11. Augmented AI (human review)
12. Real-time inference
13. Image classification
14. Forecast accuracy
15. Yield predictions

---

📊 EVALUATION METRICS CHEATSHEET

Model Performance Metrics — "ROGUE BLEU BERT"

• ROUGE (summarization quality)
• BLEU (translation quality)
• BERTScore (semantic similarity)

Traditional ML Metrics

• Accuracy: Correct predictions / Total predictions
• Precision: True Positives / (True Positives + False Positives)
• Recall: True Positives / (True Positives + False Negatives)
• F1 Score: Harmonic mean of precision and recall
• AUC-ROC: Area Under Curve — Receiver Operating Characteristic

---

🛡️ SECURITY & COMPLIANCE ESSENTIALS

IAM for AI/ML — "ROLE POLICY ENCRYPT"

1. Roles (service permissions)
2. Organizations (account management)
3. Least privilege (minimal access)
4. Encryption (data protection)
5. Policies (access control)
6. Observability (monitoring)
7. Logging (audit trails)
8. Inspection (vulnerability scans)
9. Compliance (regulations)
10. Yield (audit results)
11. Encryption at rest and transit
12. Network security (VPC)
13. Certificates (SSL/TLS)
14. Responsible AI
15. Yielding compliance reports
16. Privacy protection
17. Trusted advisor recommendations

Data Governance

• Data Lineage: Track data origins (SageMaker Model Cards)
• Data Catalog: AWS Glue for metadata
• Data Quality: Validation and monitoring
• Retention: Lifecycle policies
• Residency: Regional data storage

---

🚨 CRISIS MODE MNEMONICS

Ultra-compressed for time pressure :D

THE BIG 5 SERVICES: "B-S-C-R-T"

• Bedrock = Foundation models (Claude, Titan, etc.)
• SageMaker = Custom ML (Studio, Canvas, Autopilot)
• Comprehend = Text analysis (sentiment, entities, topics)
• Rekognition = Computer vision (faces, objects, text in images)
• Transcribe = Speech to text (audio processing)

DECISION SHORTCUTS: "WHEN IN DOUBT"

• Changing data frequently? → RAG (not fine-tuning)
• Need transparency/explainability? → PDPs, SHAP, LIME
• Cost optimization question? → Pay-per-use, auto-scaling
• Security question? → Encryption + IAM + least privilege
• Custom vs pre-built? → Custom = SageMaker, Pre-built = Bedrock

---

🎪 PRACTICE QUESTIONS & EXPLANATIONS

Domain 1 Sample Questions

Q1: Which AWS service is best for real-time sentiment analysis of customer reviews?

A) Amazon Polly
B) Amazon Comprehend
C) Amazon Textract
D) Amazon Translate

Answer: B) Amazon Comprehend

Explanation: Comprehend provides real-time natural language processing including sentiment analysis. Polly is text-to-speech, Textract extracts text from documents, Translate converts languages.

Q2: What type of learning uses labeled training data?

A) Unsupervised learning
B) Reinforcement learning
C) Supervised learning
D) Transfer learning

Answer: C) Supervised learning

Explanation: Supervised learning requires labeled data for training. Unsupervised uses unlabeled data, reinforcement uses rewards/penalties.

Domain 2 Sample Questions

Q3: What is the purpose of embeddings in generative AI?

A) To compress images
B) To convert text into numerical vectors
C) To encrypt data
D) To reduce model size

Answer: B) To convert text into numerical vectors

Explanation: Embeddings transform text into high-dimensional numerical vectors that capture semantic meaning for AI processing.

Q4: Which inference parameter controls randomness in model responses?

A) Top-k
B) Temperature
C) Top-p
D) Max tokens

Answer: B) Temperature

Explanation: Temperature controls randomness — lower values make responses more deterministic, higher values increase creativity/randomness.

Domain 3 Sample Questions

Q5: What is RAG (Retrieval Augmented Generation)?

A) A type of neural network
B) A method to combine external knowledge with LLMs
C) A security protocol
D) A data preprocessing technique

Answer: B) A method to combine external knowledge with LLMs

Explanation: RAG retrieves relevant information from external sources to augment the model's knowledge for more accurate responses.

Q6: Which AWS service is best for storing vector embeddings?

A) Amazon S3
B) Amazon DynamoDB
C) Amazon OpenSearch Service
D) Amazon CloudWatch

Answer: C) Amazon OpenSearch Service

Explanation: OpenSearch Service supports vector search capabilities for storing and querying embeddings efficiently.

---

🎯 EXAM STRATEGY TIPS

Time Management

• 65 questions in 90 minutes = ~1.4 minutes per question
• Flag difficult questions and return later
• Read case studies carefully — multiple questions per scenario
• Don't overthink — first instinct often correct

Question Approach

1. Read the question completely
2. Identify key AWS services mentioned
3. Eliminate obviously wrong answers
4. Look for AWS-specific terminology
5. Choose the most AWS-native solution

📖 RECOMMENDED STUDY RESOURCES

Official AWS Resources

1. AWS AI Practitioner Exam Guide (Primary source)
2. AWS Training and Certification portal
3. AWS Documentation for each service
4. AWS Whitepapers on AI/ML best practices

Practice Tests (Prioritized)

1. Udemy by Stephane Maarek — 260 quality questions
2. ExamTopics — Free questions with community discussions

Free Practice Resources

• AWS Skill Builder — Free digital training
• PartyRock — Hands-on generative AI playground
• Medium articles — Domain-specific practice questions

---

Good luck on your certification journey! 🎓

---