End-to-end ML systems, NGS pipelines, health informatics tools, and production data platforms.
A DNA-analysis pipeline that reads raw genetic sequencing data and predicts which population group a sample most likely came from, the same type of workflow used in ancestry testing and precision medicine. Built with industry-standard tools (BWA-MEM, GATK) and a Random Forest classifier reaching 79% accuracy and 0.96 AUC.
A deep learning model that watches wearable sensor data and predicts when a Parkinson's patient is about to "freeze" mid-step, a dangerous symptom that causes falls. Converts motion signals into images a neural network can read, reaching 0.85 AUC (vs. 0.83 for a traditional ML baseline).
A desktop app for education platforms that locks down who can sign up (invite-only, no open self-registration) and routes each user to the right dashboard by role. A companion Python layer flags suspicious logins automatically using anomaly detection, the same technique fraud-detection systems use.
A clinical decision-support tool that asks patients a questionnaire and matches them to the contraceptive resources most relevant to their situation, including transgender and gender-nonconforming (TGNC) patients, an underserved group most existing tools ignore. Built on a HIPAA-aligned database and API, with privacy protections baked in from the start.
Built a clinical ML platform for continuous glucose monitor (CGM) data that identifies hypoglycemia risk, explains predictions with SHAP, supports FHIR R4 interoperability (so it can plug into hospital record systems), and includes a validated 49-test data pipeline. Live demo available.