Machine Learning Engineer at HealthTech Inc., building predictive models for healthcare applications.
Core Work
- Developed patient risk assessment models using electronic health records (EHR) data
- Implemented federated learning techniques to train models across multiple hospital systems without centralizing sensitive patient data
- Built real-time inference pipelines for clinical decision support
Privacy-First Architecture
HIPAA compliance was non-negotiable. We implemented a federated learning architecture where model weights — not patient data — are shared between institutions. Differential privacy guarantees ensure individual patient records cannot be reconstructed from model parameters.
Technical Details
Models built with TensorFlow and PyTorch, deployed on HIPAA-compliant cloud infrastructure. The prediction pipeline processes EHR data in real-time, providing risk scores to clinicians within 200ms of record update.
Results
The risk assessment model achieved 0.89 AUROC on a held-out test set across three hospital systems, outperforming the existing rule-based screening by 23%.