Exam Overview
Exam Details
- Duration: 180 minutes
- Questions: 65
- Passing Score: 750/1000
- Format: Multiple choice & multiple response
- Cost: ~$150-300 USD
- Validity: 3 years
Exam Domains
| Domain | Weight |
|---|---|
| Data Engineering | 24% |
| Exploratory Data Analysis | 26% |
| Modeling | 36% |
| ML Implementation & Operations | 14% |
AWS ML Services Overview
| Service | Category | Use Case |
|---|---|---|
| SageMaker | Full ML Platform | Build, train, deploy any ML model end-to-end |
| Rekognition | Vision AI | Image/video: objects, faces, text, content moderation |
| Comprehend | NLP | Sentiment, entities, key phrases, language detection |
| Textract | Document AI | Extract text and data from scanned forms and documents |
| Transcribe | Speech-to-Text | Automatic speech recognition; call center transcription |
| Polly | Text-to-Speech | Convert text to lifelike speech; SSML support |
| Lex | Conversational AI | Chatbots and voice interfaces (same tech as Alexa) |
| Forecast | Time Series | ML-based demand/inventory forecasting |
| Personalize | Recommendations | Real-time personalized recommendations |
| Bedrock | Generative AI | Foundation models via API (Claude, Llama, Titan, Stable Diffusion) |
SageMaker Deep Dive
| Component | Purpose |
|---|---|
| Studio | Web-based IDE for complete ML lifecycle |
| Ground Truth | Data labeling with human reviewers + ML auto-labeling |
| Data Wrangler | Data prep and transformation; 300+ transforms; no-code |
| Feature Store | Centralized features; online (real-time) + offline (batch) |
| Training Jobs | Managed training; spot training for 70% cost savings |
| Automatic Model Tuning | Hyperparameter optimization (Bayesian/random/grid) |
| Endpoints (Real-time) | Low-latency inference; auto-scaling; multi-model endpoints |
| Batch Transform | Offline bulk inference on S3 data |
| Pipelines | MLOps CI/CD for ML workflows |
| Clarify | Detect bias and explain model predictions |
| Model Monitor | Detect data drift and model quality degradation in production |
Built-in Algorithms
| Algorithm | Problem Type | Use Case |
|---|---|---|
| XGBoost | Classification/Regression | Tabular data; most popular for structured data |
| Linear Learner | Classification/Regression | Large sparse datasets; fast training |
| K-Means | Clustering | Customer segmentation |
| BlazingText | NLP | Text classification; word embeddings |
| DeepAR | Time Series Forecasting | Multiple related time series |
| Random Cut Forest | Anomaly Detection | Detect anomalies in streaming data |
| Image Classification | Vision | Classify images (ResNet) |
📋 Study Checklist
Progress0%
- Know all AWS AI/ML managed services and their use cases
- Understand SageMaker components: Studio, Data Wrangler, Feature Store
- Know SageMaker built-in algorithms and when to use each
- Explain SageMaker training: CPU vs GPU instance types
- Know SageMaker Spot Training for cost savings
- Design end-to-end ML pipeline with SageMaker Pipelines
- Understand hyperparameter tuning strategies
- Know SageMaker Model Monitor for drift detection
- Understand A/B testing and shadow deployment for models
- Know data formats: CSV, RecordIO-Protobuf, Parquet, Pipe Mode
- Design data pipelines with Glue, Athena, Kinesis for ML
- Understand bias detection with SageMaker Clarify
- Know Rekognition: image analysis, facial analysis, content moderation
- Understand Comprehend for NLP tasks
- Know when to use Forecast vs DeepAR
- Understand Bedrock: foundation models, RAG, Agents
- Know SageMaker endpoints: real-time, serverless, batch, async
- Understand Feature Store: online vs offline store
- Design cost-optimized ML with managed spot training
- Know evaluation metrics: accuracy, F1, AUC-ROC, RMSE