Selected work

Projects in data engineering, machine learning, and applied software. For each one I've separated what I built from what a team built, and labeled prototypes and work in progress as what they are.

Projects

Greenopia logo Greenopia ToolsPython, Pandas, Firebase, React, Sheets

Greenopia data platform & app

Purpose. Give Greenopia a clean, trustworthy foundation of business data so its app can help people find sustainable places with confidence.

My contribution. I reorganized a database of more than 300,000 businesses, built scraping and enrichment pipelines over more than 30,000 records, enriched more than 50,000 businesses for the production view, and built an ingestion API and an automated path that moved more than 400 businesses into the mobile app. I also built internal tools for managing customer data.

Status. The Greenopia discovery app is live.

Full case study Live app
LA Hacks 2026

Polaris — hospital alert routing

Purpose. Get the right hospital alert to the right clinician by weighing priority, proximity, availability, and workload, with typed or voice input and separate operator and clinician views.

My contribution. Built with a team at LA Hacks 2026. Public sources don't break out individual roles, so I describe this as the team's work. Devpost lists Polaris as the Best Social Impact Hack winner.

Status. Prototype built against mocked records, not deployed clinical software.

More on Polaris Devpost
Research ToolsPython, scikit-learn, XGBoost, LightGBM, FastAPI, React

Sex-stratified statin toxicity prediction

Purpose. Estimate the risk of adverse statin reactions in female patients, a group under-represented in a lot of cardiovascular modeling.

My contribution. Built a stacked ensemble (XGBoost, LightGBM, random forest, and logistic regression with a meta-learner) on federal adverse-event reports filtered to female statin patients, with a FastAPI backend and a React interface. Started at In Transcription @ UCLA's 2026 BioHackathon.

Status. Research prototype, not a clinical recommendation tool.

Full write-up Code
Ongoing ToolsPython, XGBoost

Genetic variant classifier

Purpose. Classify genetic variants as pathogenic or not from their biological features.

My contribution. Built an XGBoost classifier on ClinVar variants (GRCh38), using LASSO and PCA to study the feature space. The repository reports 82.36% accuracy on a held-out split with five-fold cross-validation.

Status. Ongoing. Not intended for clinical use.

Code
Prototype ToolsPyTorch, FastAPI, React

BR.AI.N — medical imaging interface

Purpose. A toolkit for working with brain-tumor imaging, built around a trained segmentation model behind a dashboard.

My contribution. Combined a U-Net segmentation model with a FastAPI service and a React dashboard. Designed for clinicians to explore; not used in clinical settings.

Status. Work in progress.

Code
UCLA Data Science Union logo Data Science Union ToolsPython, Pandas

Michelin Connected Fleet braking intelligence

Purpose. Help a fleet-telematics pipeline report braking performance more usefully to drivers, fleet managers, and maintenance teams.

My contribution. Built the component that dynamically estimates how far braking-score thresholds could be lowered before reporting to drivers and fleet teams.

Data Science Union
Bruin Sports Analytics logo Writing ToolsPython

Formula 1 pit-stop analysis

Purpose. Look at what actually drives pit-stop times in Formula 1.

My contribution. Wrote and published the analysis with Bruin Sports Analytics. (The original article link has since gone offline.)

Bruin Sports Analytics
Side project ToolsPython

Daily news agent

Purpose. A small agent that assembles a daily news digest.

My contribution. Built it end to end as a personal exploration of agent workflows.

Code