MLOps Engineer 2026

MLOps Pipeline — SageMaker

End-to-end MLOps pipeline on AWS SageMaker: automated preprocessing, training, evaluation, conditional model registration, and Lambda-triggered retraining on drift detection. Built with the SageMaker Pipelines SDK and Terraform infrastructure.

SageMaker PipelinesPythonLambdaS3TerraformCloudWatch
Source Code

Production-grade MLOps pipeline that automates the full model lifecycle from raw data to registered model, with automated retraining triggered by data drift.

What it does

  • 5-phase SageMaker Pipeline: Preprocess → Train → Evaluate → Conditional Register → Baseline
  • Model only registered if ROC-AUC exceeds threshold — no manual promotion gate needed
  • Lambda function monitors CloudWatch drift metrics and re-triggers the pipeline automatically
  • 24 unit tests (TDD) covering preprocessing, drift detection, and Lambda trigger logic

Technical highlights

  • PropertyFile + JsonGet condition for automated quality gate
  • numpy.bool_ JSON serialisation fix (cast to Python bool before json.dumps)
  • KS test for numeric drift, chi-squared for categorical drift
  • Terraform-managed SageMaker execution roles, S3 buckets, and Lambda trigger