Optum

Problem

Healthcare data is fragmented. Clinical records, claims, pharmacy data, and lab results live in separate systems. Manual review dominates: nurses document by hand, claims require human approval, and care gaps go undetected until patients show up in the ER. The challenge is building ML systems that run reliably at this scale while meeting HIPAA and SOC 2 requirements.

Approach

• 01Own end-to-end ML pipeline development: data ingestion, feature engineering, model training, deployment, and monitoring
• 02Deploy models to production on AWS with Kubernetes orchestration and automated rollback on performance degradation
• 03Build data processing jobs handling millions of healthcare records with strict SLA requirements
• 04Implement model observability: drift detection, prediction confidence tracking, and alerting for production anomalies
• 05Work with data scientists to translate research models into production-grade services with latency and throughput guarantees
• 06Maintain HIPAA compliance across all data pipelines with encryption, access controls, and audit logging

Design Decisions

Technology Stack