My Work
Things I have built at the edge of machine learning and medicine, all aimed at making healthcare earlier, cheaper and easier to reach.
Research
Neuroscan360
A multimodal early detection system for neurodegenerative diseases, built around a 23-game suite that tests memory, motor coordination, speech and vision.
Role: All of the software (hardware by my collaborator)
Case study →Prostate cancer detection at NYU Langone
Multimodal deep learning for prostate cancer detection, so fewer patients go through biopsies they did not need.
Role: Research Engineer Intern
Case study →Lupus, gene expression and fairness
A class project asking whether gene expression can detect lupus and estimate how severe it is, and whether those models treat patients of every race and sex fairly.
Role: Disease progression models (with Rachel Yuan, who did the fairness evaluation)
Case study →CurioCity
An app that helps groups of friends decide where to go in Los Angeles, with suggestions based on distance, budget, transportation and activity preferences. I formed and led a 14-person team of engineers, UX researchers and designers, ran user research with UCLA students, and we built an MVP and began a group-match feature.
Role: Formed and led a 14-person team
Case study →Markov chain models of cichlid fish behavior
I built Markov chain models to find and predict behavioral patterns in cichlid fish, starting with a study of how stressful events affect male sexual performance and reproduction. I taught myself Markov chains in Python, then rewrote the code to be much shorter and general enough to analyze behavior in any animal species.
Case study →Alzheimer’s classification from MRI scans
Neural network image classifiers for Alzheimer’s disease, trained on 6,400 MRI scans from Kaggle. A ResNet 101 reached 63% accuracy, and a custom eight-layer CNN reached about 97% at telling demented from non-demented scans. I also tested the CNN on edited images, to see what it had actually learned.
Case study →Experience
Machine Learning Fellow at ChartR
This past summer, I built pipelines to evaluate clinical decision-support agents against real reports, to measure how well they actually perform. I can’t share the details publicly, but I’m happy to talk about it.
Software Engineering Intern at Infosys BPM
I built a scheduled Python system that collects GDP data for 190 countries with Playwright, builds Excel reports, and emails them through Outlook, saving about five hours a week. Runs succeed over 95% of the time, and standardizing country names cut formatting fixes by about 70%.
Applied Data Science Intern at Impetus
At a healthcare data startup, I streamlined data pipelines to cut model training time by 20%, and reached 94% model accuracy through optimization and feature engineering. Remote.
Skills: Python, SQL
Product Management Intern at Nymble
I led cross-functional workshops that shaped the roadmap and decreased user churn by 27%, drove a 21% increase in conversions through Bay Area campaigns and demos, and grew brand reach by 35% across web and social. San Mateo, California.
Skills: Go-to-market strategy, JavaScript
Projects
Can I Watch It At Night?
A movie recommendation app that tells you whether a title suits night-time viewing, based on your threshold for scary. It uses a custom database of 9,500+ titles, vector embeddings, and a TF-IDF backend that responds in under 1.5 seconds.
Case study →Autonomous job application agent
An agent built on the Anthropic SDK that finds roles, judges my fit, researches the company, drafts outreach, and fills out applications on real job portals, using 55 tools across five namespaces.
Case study →Spotify playlist feature analysis
A Python script that pulls song data for any Spotify playlist and plots danceability, energy and valence with seaborn. I tried it on a playlist of 100 songs with LA in the title, and they came out energetic and positive, just like the city.
Safr
A concept app that makes mental health care easier to reach, by collecting providers on one platform, showing affordable options, and redesigning intimidating clinical questionnaires. I designed it from my own experiences, and those of people close to me, to help reduce the stigma around seeking therapy.