ML / MLOps Engineer — building end-to-end machine learning systems
I design and build ML systems end-to-end — from data ingestion and model training to deployment, serving, and monitoring. Most of my recent work sits at the intersection of clinical/healthcare ML and production MLOps infrastructure on AWS.
CTN-0094 Pipeline — a reproducible ETL and analysis pipeline for the CTN-0094 substance use disorder treatment dataset. Uses propensity score matching to study algorithmic fairness in treatment outcome prediction. Co-published in CDAR (Wiley, March 2025).
Focused practice across the pieces that make up a real SageMaker pipeline, using a "bring your own container" scikit-learn pattern:
- Training Job — custom container training, hyperparameters, artifact output to S3
- Processing Job — scalable preprocessing/feature engineering jobs decoupled from training
- Batch Transform — offline/batch inference on large datasets without a live endpoint
- Endpoint — real-time inference hosting and invocation
- SageMaker Pipelines — wiring the above into a single reproducible, orchestrated workflow
| Project | What it demonstrates |
|---|---|
| Meridian — churn intelligence platform | XGBoost/LightGBM model, MLflow + DVC tracking, FastAPI serving, Evidently AI monitoring, React/Vite frontend with SHAP explanations, deployed on Fly.io via GitHub Actions |
| NYISO energy demand forecasting | Full ML lifecycle: event-driven ingestion from EIA + NOAA APIs, feature engineering, SageMaker Pipelines orchestration, MLflow model comparison/promotion |
| Insurance claims fraud detection | Two parallel infra tracks on the same use case — SageMaker-native (Training Jobs, Endpoints, Pipelines) vs. containerized (Docker, MLflow, FastAPI + Gradio, ECS Fargate + ALB, CI/CD) |
(Add repo links here once each is public — happy to help write project-specific READMEs too.)
- Building out AWS/SageMaker-native MLOps pipelines end-to-end (ingestion → training → deployment → monitoring)
- Clinical ML pipeline research for CTN-0094 (opioid use disorder treatment outcomes)
- Comparing SageMaker-native vs. containerized (Docker/ECS) deployment patterns for the same model
Previously: Software Engineer/SOC @ NYPD · Now: ML/MLOps research @ CCNY