Case Studies
End-to-end ML systems built and shipped in production.
Medical Document Classifier
How do you route 50K+ medical documents automatically with near-zero error?
Built a production document classification system using Longformer transformers to automatically categorize and route high-volume medical documents. The system processes tens of thousands of records with near-zero misclassification, replacing a costly manual triage workflow.
Cancer Risk Prediction Pipeline
Can CMS Medicare claims data predict which patients are highest risk?
Designed and deployed a multi-stage ML pipeline on Databricks using CMS Medicare claims data to predict cancer treatment pathways and flag high-risk patients. The Delta Lake medallion architecture ensures data quality across the pipeline, and MLflow tracks all experiments.
Phishing Detection Under Adversarial Attacks
Do ML phishing classifiers survive Unicode homoglyph obfuscation?
Investigated the robustness of ML-based phishing classifiers when subjected to Unicode homoglyph substitution attacks. Developed augmented training data with adversarial examples to harden the classifier and published findings with a public dashboard.
DHS Vaccination Coverage Analysis
What do DHS surveys reveal about vaccination coverage gaps across countries?
Analyzed Demographic and Health Surveys (DHS) data across multiple countries to surface disparities in childhood vaccination coverage. Identified socioeconomic and geographic correlates of under-vaccination, informing federated learning research for health equity.