Skip to content
SDB

Cloud Technology

AWS IoT & AI Reference Architecture

The broadest industrial IoT and AI portfolio — from Greengrass and SiteWise at the edge to SageMaker MLOps, Bedrock foundation models, and agentic workflows.

Amazon Web Services · 5 architecture layers

End-to-end data flow

01

Sensors & PLCs

OPC-UA / MQTT

02

Greengrass Edge

Local inference

03

IoT Core / Kinesis

Ingestion broker

04

S3 / Timestream

Time-series store

05

SageMaker / Bedrock

ML + GenAI + Agents

06

QuickSight / Apps

Dashboards & APIs

Architecture layers

L1

Device & Edge Layer

AWS IoT Greengrass

Run local Lambda, containers, and ML models at the edge; sync securely with AWS

AWS IoT SiteWise Edge

Collect and process industrial data at source from OPC-UA servers

AWS IoT FleetWise

Vehicle and machinery telemetry at scale with custom signal decoders

AWS Panorama

Deploy computer-vision models to industrial cameras and on-prem appliances

AWS IoT Device Defender

Audit device fleets, detect anomalous behaviour, and enforce security policies

L2

Connectivity & Ingestion Layer

AWS IoT Core

Managed MQTT/HTTPS broker — connect billions of devices with device shadows

Amazon Kinesis Data Streams

Real-time streaming ingestion for high-throughput sensor telemetry

Amazon Data Firehose

Serverless delivery of streams into S3, Redshift, OpenSearch, or Splunk

AWS IoT Events

Detect and respond to equipment state changes and multi-sensor fault patterns

L3

Processing & Storage Layer

AWS IoT TwinMaker

Compose digital twins linking real-world data with 3D scenes

Amazon S3

Petabyte-scale object storage — foundation of IoT data lakes on AWS

Amazon Timestream

Purpose-built time-series database for IoT telemetry

Amazon DynamoDB

Sub-millisecond NoSQL for device state, shadows, and real-time lookups

L4

AI / ML Layer

Amazon SageMaker

End-to-end MLOps — labeling, training, registry, endpoints, and monitoring

Amazon SageMaker Unified Studio

Unified workspace for data, analytics, and AI teams on a shared lakehouse

Amazon Bedrock

Serverless access to Claude, Llama, Mistral, Amazon Nova and Titan models

Amazon Bedrock Agents

Orchestrate tools, knowledge bases, and multi-step agent workflows for OT/IT tasks

Amazon Q

Generative assistant for business and developers grounded in enterprise data

Amazon Lookout for Equipment

Detect industrial equipment anomalies using models trained on your sensor data

Amazon Monitron

End-to-end predictive maintenance — wireless sensors plus ML anomaly detection

Amazon Rekognition

Vision APIs for PPE detection, defect inspection, and object analysis

L5

Application & Visualization Layer

Amazon QuickSight

BI dashboards for IoT KPIs with auto-scaling concurrency

AWS AppSync

Real-time GraphQL APIs for IoT dashboards and mobile apps

AWS IoT SiteWise Monitor

No-code industrial asset dashboards without custom development

Amazon Managed Grafana

Grafana with native AWS data sources and IAM authentication

Industrial AI use cases

Predictive Maintenance

SageMaker + Lookout for Equipment + Monitron

Visual Quality Inspection

Rekognition + Panorama + SageMaker Edge

Conversational & Agentic AI

Bedrock Agents + Amazon Q + Knowledge Bases

Demand Forecasting

Amazon Forecast + SageMaker DeepAR / Canvas

Anomaly Detection

SageMaker + Lookout for Metrics + IoT Events

Digital Twin Intelligence

IoT TwinMaker + SageMaker + Grafana

Key differentiators

Deepest IoT portfolio

SiteWise, TwinMaker, Monitron, Panorama, FleetWise and Device Defender — purpose-built industrial building blocks.

Greengrass at the edge

Run Lambda, containers and SageMaker models locally from Raspberry Pi to industrial gateways.

Bedrock GenAI & Agents

Single API for foundation models plus agents, knowledge bases and guardrails — no GPU fleets to manage.

SageMaker MLOps

From Ground Truth labeling to feature store, registry, A/B testing and drift detection in one platform.