AWS AI/ML Service Landscape

1. Three Layers of AWS AI/ML

AWS AI/ML services are organized in three layers based on the level of ML expertise required.

AWS AI/ML Stack:

Layer 3: AI SERVICES (no ML expertise needed)
  Pre-trained APIs. Call via SDK. No data, no training.
  Rekognition, Textract, Comprehend, Translate, Polly, Transcribe, etc.

Layer 2: GENERATIVE AI (prompt engineering + customization)
  Foundation models via API. Optional RAG/fine-tuning.
  Amazon Bedrock, Amazon Q, PartyRock

Layer 1: ML PLATFORM (full ML expertise)
  Build, train, deploy custom models. Full control.
  Amazon SageMaker

2. Amazon Bedrock (Generative AI)

What is Bedrock? Amazon Bedrock is a fully managed service to access foundation models from multiple providers via a single API. No infrastructure to manage. Models from: Anthropic (Claude), Meta (Llama), Amazon (Titan), AI21 Labs, Cohere, Stability AI, Mistral. Use for: text generation, summarization, chat, code, and image generation.

Bedrock Key Features

1. Single API to access multiple foundation models (compare and switch easily)
2. Serverless: no GPU instances to manage. Pay per token (input + output).
3. Knowledge Bases: built-in RAG. Connect your S3 documents. Auto-creates embeddings + vector store.
4. Agents: build AI agents that can take actions (call APIs, query databases, execute tasks)
5. Guardrails: filter harmful content, PII, and off-topic responses. Safety controls.
6. Custom Models: fine-tune FMs on your data within Bedrock
7. Model evaluation: compare model outputs for your specific use case
8. Data privacy: your data is NOT used to train base models. Stays in your account.

Bedrock Models

3. Amazon Q (AI Assistant)

1. AI-powered assistant for different audiences
2. Amazon Q Business: answer questions from company data (documents, wikis, Slack, SharePoint, S3)
3. Amazon Q Developer: AI coding assistant (code suggestions, debugging, transformation, security scanning)
4. Amazon Q in QuickSight: natural language BI queries and dashboard generation
5. Amazon Q in Connect: AI-powered contact center agent assist
6. Uses RAG automatically (connects to your enterprise data sources)

4. Amazon SageMaker (ML Platform)

What is SageMaker? SageMaker is a fully managed ML platform for building, training, and deploying custom ML models. It covers the entire ML workflow: data labeling, notebooks, training, hyperparameter tuning, deployment, and monitoring. Use SageMaker when pre-built AI services don’t meet your needs, and you need full custom model control.

SageMaker Key Components

5. Pre-Trained AI Services (No ML Expertise)

6. AI/ML Service Selection Guide

Exam Tip AI/ML Services: Match the use case to the service. "GenAI/foundation models" = Bedrock. "Custom ML" = SageMaker. "Images/video" = Rekognition. "OCR/documents" = Textract. "Sentiment" = Comprehend. "Translate" = Translate. "Text→Speech" = Polly. "Speech→Text" = Transcribe. "Chatbot" = Lex. "Search" = Kendra. "Recommendations" = Personalize. "Forecast" = Forecast. "Fraud" = Fraud Detector. "No-code ML" = Canvas. "RAG on your docs" = Bedrock Knowledge Bases.