AWS Cloud Architecture Design
End-to-end AWS architecture for a real-time ML recommendation app at a gas plant.
Scaled to 300% more users with no performance loss.
Architecture diagram · zoom, drag or double-tap to explore
100%
Follow the data · tap a step
Problem
Plant process data lived in OSIsoft PI, and the business needed ML recommendations in real time — served with the same rigour as the plant's own engineering systems.
Approach
- Ingested PI-BA and PI-AF tags onto AWS EC2, then ran the ETL process into RDS (SQL Server).
- Served the models as FastAPI services on ECS — one path for on-demand client requests, one for scheduled ML output.
- Triggered scheduled runs with EventBridge and Lambda; .NET Core backend and React frontend on AWS, with Azure AD handling authentication and authorisation.
- Shipped through Azure DevOps CI/CD with Katalon QA automation and Acunetix security assessment; monitored with CloudWatch and Google Analytics.
Results
- Scaled to 300% more users with no performance loss.
- One architecture covering ingestion, ML serving, the app and CI/CD.
Industry
Energy
Tech stack
AWSEC2ECSLambdaRDSFastAPI.NET CoreReactAzure DevOps