Forward-deployed to a $31B health system to bring real-time eligibility into live revenue-cycle operations.

• Delivered $13M in error reduction on a real-time eligibility implementation for a $31B health-system client, advancing a decision that had stalled in discussion for 18 months.

Own production ML across the largest publicly operated health plan in the U.S., from ETL to models validated in the real world.

• Sole contributor on a claims-audit application for claim adjudication operations, from PySpark distributed pipelines and ACID transactional storage to delivery for a 12-person audit team, driving operational improvement and efficiency through root-cause issue identification and resolution, with $10M in error reduction validated by a randomized controlled trial.
• Built and shipped production Python REST ChatGPT API services on Azure (Docker, monitoring, anomaly detection) with agentic Chain-of-Thought and CoVe (Chain-of-Verification) hallucination control, improving medication identification for members and preventing $2M in adverse outcomes.
• Core designer of the DHCS RSST geospatial health-improvement scoring module, built with state entities to surface equity improvements across California and deployed to a 55+ person analytics team.
• Scoped a regulatory-compliant intervention system in Python/PySpark with a 17-person PharmD team, running 16 campaigns across population-health measures, diabetes glucose control, colorectal cancer screening, and blood pressure, that beat competitor benchmarks.
• Applied XGBoost, CatBoost, and LightGBM bagging and stacking with post-training equity adjustment to target population-health measures (e.g., Diabetes Glucose Control) across 2.7 million members.
• Built and maintained GitHub CI/CD pipelines with automated tests and safe deploys across SQL, Snowflake, and PySpark.

Ran the compute and research pipeline behind two published transformer models for air-pollution forecasting.

• Designed and managed a Weights & Biases-integrated GCP environment for high-compute PyTorch Lightning models, producing two air-pollution-prediction publications using state-of-the-art transformers.
• Taught graduate students algorithmic proofs and implementation in Java.