AWS for a Health-AI App (Terraform)

Context

A Terraform-based AWS setup built for the engineering team of a health-AI startup. It is built on Terraform, AWS ECS Fargate, Docker, and GitHub Actions, with 9 modules, dev and prod environments, auto-scaling services, CI/CD pipelines, and handover documentation.

The Hard Part

The client had a medical-AI application but no production cloud environment and no way to operate it independently. The work covered a cloud environment, containers, a database, networking and security, monitoring, CI/CD, and handover documents, and the client's team needed to deploy and run the app without ongoing dependency.

What I Built

• 9 Terraform modules for networking, security, IAM, secrets, database, cache, storage, compute, and monitoring, wired in dependency order for dev and prod environments
• ECS Fargate setup for 4 services (frontend, clinical, image, and worker) with CPU and memory limits, service discovery, and per-service log groups with 7-day retention
• Application load balancer with path-based routing to the clinical, image, and frontend services, plus CPU and memory auto-scaling for each service
• Sizing for each environment: dev with a small single-AZ database and no deletion protection, prod with a medium multi-AZ database and deletion protection
• 2 GitHub Actions pipelines: one for Terraform format, validate, plan, and apply through a production GitHub environment, and one that builds 4 Docker images, pushes them to ECR, and deploys to ECS, with AWS login through OIDC and no stored keys
• Security and monitoring modules for IAM, security groups, secrets, CloudTrail, CloudWatch, SNS, and budgets
• 4 Dockerfiles with multi-stage builds, non-root users, and health checks

The Build

Measured Results

Tech Stack