Amazon’s AI stack, centered around Amazon Web Services (AWS), is vertically integrated. It spans from raw infrastructure all the way to end-user AI applications. The key idea: AWS doesn’t just give you models; it gives you every layer needed to build, train, deploy, and scale AI systems in production.
Let’s break it down cleanly, layer by layer.
Layer 1: Data Layer
Purpose: Enables data collection, storage, processing, and preparation for ML
- Amazon S3: Scalable object storage for datasets and model artifacts
- Amazon Redshift: Data warehousing for analytics and ML training
- AWS Glue: Serverless data integration and ETL (Extract, Transform, Load) service
- Amazon Kinesis: Real-time data streaming for ML applications
- AWS Lake Formation: Build, secure, and manage data lakes for ML
Layer 2: Machine Learning Platform Layer (Bedrock)
Purpose: Hosts Amazon’s in-house Foundation models as well as APIs to access partners’ models
· Amazon Foundation Models: Amazon has Nova family of Foundation models to cater to a variety of needs
o Nova Pro: Designed for complex, multi-step tasks, offering top-tier performance for reasoning and understanding
o Nova Lite: An efficient, fast, and cost-effective model suitable for text, image, and video tasks
o Nova Micro: An extremely fast and lightweight model focused on high-throughput, low-latency text tasks
o Nova 2 Omni (Preview): A multimodal reasoning model capable of processing text, images, video, and speech, while natively generating text and images
o Nova Reel: Dedicated to generating short video content, including the ability to take reference images to guide video creation
o Nova Canvas: Generates images and offers editing capabilities
o Nova Act: Foundation model that interacts with UIs: clicks buttons, fills forms, navigates apps - ideal for legacy system automation
- AWS IoT Greengrass: Run ML models locally on edge devices
- AWS Panorama: Computer vision at the edge
- AWS Outposts: Run AWS infrastructure on-premises for low-latency ML
Layer 3a: Tools for Builders’ Layer
Purpose: Tools for building custom ML models and workflows
- SageMaker Studio: Integrated development environment (IDE) for ML
- SageMaker Autopilot: Automated model building
- SageMaker JumpStart: Pre-built models and solutions
- SageMaker Pipelines: Orchestrate ML workflows
- SageMaker Feature Store: Central repository for ML features
- SageMaker Clarify: Detect bias and explain model predictions
- SageMaker Edge Manager: Deploy models to edge devices
- AWS Deep Learning AMIs: Pre-configured environments for deep learning frameworks (TensorFlow, PyTorch, etc.)
- AWS Deep Learning Containers: Docker images for deep learning
Layer 3b: AI Application Layer
Purpose: Provide pre-build plug-n-play AI APIs for ML workloads
- Amazon Rekognition: Image and video analysis (e.g., object detection, facial recognition)
- Amazon Polly: Text-to-speech service
- Amazon Lex: Build conversational interfaces (chatbots, voice assistants)
- Amazon Comprehend: Natural language processing (NLP) for text analysis
- Amazon Forecast: Time-series forecasting
- Amazon Personalize: Real-time personalized recommendations
- Amazon Textract: Extract text and data from documents
- Amazon Transcribe: Automatic speech recognition
- Amazon Translate: Language translation
- Amazon Forecast: Time-series forecasting for demand planning, inventory optimization, etc.
- Amazon Fraud Detector: Fraud prevention for transaction risk scoring, account takeover detection
- Q Developer: Assistant for developers, acts as a pair programmer, helping write, debug, and upgrade code (including complex tasks like migrating legacy Java code). Amongst its capabilities include coding suggestions and security scanning.
- Q Business: Ingests data from 40+ systems like Amazon S3 and Salesforce to help build “Q Apps” for sales, lawyers etc. to perform Q&A to help users get answers to their questions, provide summaries, generate content, and securely complete tasks based on data and information in their enterprise systems.
Users can also use Amazon Q Apps to generate apps in a single step from their conversation with or by describing their requirements.
- Q in QuickSight: Natural language BI for asking questions about data
Which Layer Should You Use?
|
Requirement |
Layer |
Key Service |
|
I want to build a custom LLM from scratch |
Tools for Builders’ Layer |
SageMaker |
|
I want to build an AI agent for my app |
Machine Learning Platform Layer |
Bedrock |
|
I need an AI to help my employees work faster |
AI Applications Layer |
Amazon Q |
Posts
