AWS AI/ML Services Overview

AWS organizes its AI and ML offerings into three layers. Picking the right layer is primarily about how much of the ML stack you want to own: more managed at the top, more flexibility at the bottom.

Layer 1 — Generative AI (Foundation Models)

• Amazon Bedrock: Managed API for foundation models from Anthropic, Meta, Mistral, Cohere, AI21, Stability, and Amazon — plus Knowledge Bases (RAG), Agents, and Guardrails.
• Amazon Q: AWS-native AI assistant tuned for business data (Q Business) and AWS development (Q Developer) — built on Bedrock under the hood.
• SageMaker JumpStart: Deploy and fine-tune open-source foundation models (Llama, Mistral, Falcon, Stable Diffusion) on dedicated SageMaker endpoints.

Layer 2 — Task-Specific AI APIs

• Amazon Comprehend / Comprehend Medical: NLP — entities, sentiment, PII redaction, clinical concept extraction.
• Amazon Textract: Document AI — OCR, forms, tables, Analyze Expense / ID / Lending.
• Amazon Rekognition: Computer vision — labels, faces, moderation, custom image classifiers.
• Amazon Transcribe & Polly: Speech-to-text and text-to-speech; Transcribe Call Analytics surfaces conversation insights.
• Amazon Translate: Neural machine translation across 75+ languages.
• Amazon Personalize: Managed recommendation and personalization models trained on your interaction data.
• Amazon Forecast: Time-series forecasting using statistical + deep-learning ensembles.
• Amazon Kendra: Enterprise semantic search across connectors to SharePoint, Confluence, S3, and more.

Layer 3 — ML Platform

• Amazon SageMaker: Full ML lifecycle — labeling, training, tuning, deployment, monitoring, feature store, pipelines.
• EC2 + Deep Learning AMIs: Raw compute for teams that want to manage their own training stack.
• AWS Trainium / Inferentia: Purpose-built accelerators for cost-efficient training and inference.

Choosing Between Layers:

1. Start with task-specific APIs if a managed API matches your use case — fastest to production, no training data required.
2. Use Bedrock when the task is generative, open-ended, or benefits from foundation-model reasoning. Add Knowledge Bases for RAG and Guardrails for safety.
3. Drop to SageMaker when you need full control over training, hosting, or custom model architectures.