Worked on AI systems for healthcare Utilization Management (UM) at Elevance Health, focusing on clinical criteria automation, predictive modeling, and production ML infrastructure.

Some of the stuff I did:

• Improved a claims denial risk scoring system through feature engineering, raising PR-AUC from 0.82 to 0.90, and implemented model routing to reduce inference latency by 10%.
• Integrated Fiddler across 15+ production machine learning models (primarily XGBoost and ensemble classifiers) for case-detail probability predictions to monitor data drift and model performance. Analyzed telemetry across 5 production services to identify opportunities for latency reduction, precision gains, and infrastructure cost savings.
• Contributed to an AI-assisted Utilization Management automation platform:
  • Built a configurable DAG-based workflow engine that coordinates hybrid retrieval (semantic and BM25 RAG), structured document extraction using OCR and LLMs, and deterministic clinical criteria resolution.
  • Developed a multimodal, long-context AI agent that analyzes unstructured clinical policy documents and converts them into executable YAML DAG workflows.
• Deployed and orchestrated production AI/ML microservices on Kubernetes using Kubeflow, supporting containerized model serving across a team of 13+ data scientists and engineers.