☁️

AWS AI Services

AWS fully managed AI services

⏱️ Estimated reading time: 30 minutes

Amazon SageMaker

Fully managed platform to build, train, and deploy ML models at scale.

Components



SageMaker Studio


- Integrated web IDE
- Jupyter notebooks
- Team collaboration

SageMaker Autopilot


- Automatic AutoML
- Generates models without code
- Explains model decisions

SageMaker Training


- Distributed training
- Spot instances to reduce costs
- Multiple frameworks (TensorFlow, PyTorch, etc.)

SageMaker Inference


- Real-time endpoints
- Batch transform
- Serverless inference
- Edge deployment

🎯 Key Points

  • ✓ SageMaker covers the full ML lifecycle: experimentation, training, deployment and monitoring
  • ✓ Use Autopilot for quick prototypes but validate models manually before production
  • ✓ Use spot instances to reduce training costs where possible
  • ✓ Version models and notebooks for reproducibility
  • ✓ Set up pipelines and automated tests for ML CI/CD

Computer Vision Services

Amazon Rekognition



Capabilities:
- Object and scene detection
- Facial recognition
- Celebrity analysis
- Content moderation
- Text detection in images (OCR)
- PPE detection (Personal Protective Equipment)

Use cases:
- User verification
- Security and surveillance
- Media analysis
- Content moderation

Amazon Textract



Capabilities:
- Text extraction from documents
- Form data extraction
- Table analysis
- Signature detection

Use cases:
- Invoice processing
- Document digitization
- Process automation

🎯 Key Points

  • ✓ Choose Rekognition for out-of-the-box use cases: authentication, moderation, OCR
  • ✓ Use Textract for complex data extraction and validate structured outputs
  • ✓ Consider privacy and consent when processing images and video
  • ✓ Assess cost/latency tradeoffs for real-time vs batch solutions
  • ✓ Test with real data and measure accuracy across subgroups

Natural Language Services

Amazon Comprehend



Capabilities:
- Sentiment analysis
- Entity extraction
- Language detection
- Topic modeling
- Document classification
- PII detection

Amazon Lex



Capabilities:
- Chatbot creation
- Conversational interfaces
- Speech recognition
- AWS service integration

Amazon Polly



Capabilities:
- Text-to-speech conversion
- Multiple languages and voices
- Realistic neural voices
- SSML for pronunciation control

Amazon Transcribe



Capabilities:
- Speech-to-text conversion
- Speaker identification
- Custom vocabularies
- Automatic PII redaction

🎯 Key Points

  • ✓ Comprehend provides quick text insights without training models
  • ✓ Lex is useful for dialog-driven chatbots; Bedrock/LLMs can complement NLU
  • ✓ Polly and Transcribe enable integrated speech-to-text and text-to-speech pipelines
  • ✓ Assess language quality and custom vocabularies for domain-specific use
  • ✓ Integrate PII detection and redaction when handling sensitive data

Additional Services

Amazon Translate


- Neural machine translation
- Support for 75+ languages
- Real-time translation
- Custom terminology

Amazon Personalize


- Personalized recommendations
- ML without expertise required
- Real-time integration
- Use cases: e-commerce, streaming

Amazon Forecast


- Time series predictions
- ML-based
- Multiple automatic algorithms
- Use cases: inventory, demand

Amazon Kendra


- Intelligent enterprise search
- Semantic search
- Data source connectors
- Accurate answers from documents

🎯 Key Points

  • ✓ Compare Translate, Personalize, Forecast and Kendra by purpose: translation, recommendation, forecasting or search
  • ✓ Personalize reduces time-to-market for recommendations but requires well-modeled user data
  • ✓ Forecast is advantageous for time-series with clear patterns and historical data
  • ✓ Kendra improves semantic search across enterprise document stores
  • ✓ Measure business impact: CTR, recommendation accuracy, prediction error
← Back to AWS-AIF-C01