sagemaker-iac

Overview

sagemaker-iac is a production-grade Machine Learning Platform Control Plane built on AWS using Terraform and GitOps principles.

This repository provisions and governs the infrastructure required to support:

• Continuous model training
• Model registry & lifecycle automation
• GitOps-based model deployment
• Observability-driven ML operations
• Secure, environment-isolated ML workloads

This is not a single ML project. This is reusable ML Platform-as-Code.

---

Core Objectives

• Infrastructure modularity
• Least-privilege IAM enforcement
• Environment promotion (dev → staging → prod)
• GitOps-first Kubernetes deployment (ArgoCD)
• Automated retraining pipelines
• Drift-aware model observability

---

Architecture Layers

1. Infrastructure Layer

   • VPC networking (private subnets)
   • EKS cluster
   • SageMaker domain
   • IAM boundary model
   • S3 data lake
   • ECR repositories

2. Platform Layer

   • ArgoCD bootstrap
   • Helm-based service deployment
   • MLflow stack
   • Observability stack (Prometheus + Grafana)

3. ML Lifecycle Layer

   • Event-driven retraining
   • Model evaluation gates
   • Registry-based promotion
   • Zero-downtime rollout

---

Repository Structure

terraform/ Infrastructure modules and environments platform/ GitOps bootstrap and Helm applications ml-platform/ Reusable ML lifecycle tooling observability/ Monitoring stack and dashboards cicd/ CI/CD automation definitions examples/ Reference ML workloads docs/ Architecture and operational runbooks

---

Deployment Philosophy

• Modules represent capabilities
• Environments represent composition
• No manual kubectl in production
• Infrastructure changes require review
• Model promotion is automated and gated

---

Environments

• dev
• staging
• prod

Each environment maintains isolated state and configuration.

---

Engineering Standards

• Pre-commit enforced validation
• Security scanning (Checkov)
• Terraform linting (tflint)
• Immutable artifacts
• GitOps-based workload deployment

---

Target Audience

Internal ML teams that require:

• Secure training environments
• Automated deployment pipelines
• Observability for model health
• Governance-ready infrastructure

---

Status

Platform under active development.

---

👨‍💻 Author

Ajeet Singh — DevOps & Platform Engineering https://github.com/singhajeet79