MLOps Credit Risk Pipeline

A production-grade MLOps pipeline for credit risk assessment in Fintech lending. Automates model training on Amazon SageMaker and serves real-time default-probability predictions through a FastAPI inference server running on Amazon EKS, with all AWS infrastructure managed as code using Terraform.

Key Achievements

• Infrastructure as Code: Provisioned a complete AWS environment (VPC, EKS cluster, SageMaker, ECR, S3, IAM) using Terraform with remote state in S3, enabling repeatable standup and teardown of the entire stack.
• Managed model training: Used Amazon SageMaker to run ephemeral XGBoost training jobs on the LendingClub dataset (~1.3M loan records), with model artifacts stored in S3 and decoupled from the serving layer.
• Real-time inference on EKS: Deployed a FastAPI inference server as a Kubernetes Deployment behind a LoadBalancer Service on EKS, with the container downloading its model from S3 at startup to keep the image model-version-agnostic.
• Production networking and security: Placed EKS worker nodes in private subnets with outbound traffic routed through a NAT Gateway, and enforced least-privilege IAM roles for SageMaker and EKS node access.