I use data where sport usually uses memory or prestige — who gets minutes on a school roster, and which city can actually host a Super Bowl.
Accepted · EDUAIT 2026App Store
FairPlay Lab
I built this at SoftCom Lab, Cal Poly Pomona, with Kyler Harris
I built a mobile app that turns attendance, RPE, match minutes, and pitch events into substitution advice a volunteer coach can open and audit. I kept the scoring rule-based so every suggestion names the inputs that produced it, then I audited the engine and found that RPE moved minutes the most and that a high-load threshold inverted the intended penalty. The paper was accepted at EDUAIT 2026, the app is on the App Store, and I beta-tested it on my varsity team.
My team built a 24-indicator model to rank Super Bowl host-city sustainability across carbon, energy, transport, water, waste, and lodging. Transportation and energy carried the largest weights, and we recommended Minneapolis for 2029. The paper received an international honorable mention, top 20% of HiMCM entries.
I test claims schools treat as obvious — whose writing looks “AI,” whether a better school label raises scores, whether “private” is just a resource bundle.
Accepted · The Curieux Review
AI detectors on ELL school essays
I did this study, mentored by Dr. Mani at Carnegie Mellon
I tested whether free AI writing detectors flag English-learner school essays more often than native ones, the pattern reported on TOEFL text. I scored two samples of 200 human-written ASAP 2.0 / ELLIPSE essays with ZeroGPT and an open RoBERTa classifier. ELL essays were not flagged more often; false positives tracked shorter, less lexically diverse writing instead. The paper has been accepted by The Curieux Review; the practical consequence is that a school should not treat a single detector score as misconduct.
My team — Irene Gong, Kristen Gumagay, and me — presented Cullen, Jacob & Levitt (2006)
My team walked through Chicago school-choice lotteries to separate ITT from LATE: winning a seat is random, actually enrolling is not. Winners attended objectively better schools, but test scores and graduation barely moved, while arrests and discipline fell. That null academic result is the finding I now carry into my senior project, instead of treating it as a failed design.
I am designing this as my Basis Independent Brooklyn senior research, 2026–27
I am asking whether NYC public–private gaps are explained by measurable resources rather than the sector label — a follow-on to the UChicago lottery finding that a better school name did not raise test scores. I will use public school-level files; data collection has not started.
Making numbers readable
I design charts so a non-expert can see a disparity without a methods section.
Vanderbilt PTY · 2025–26
Mental-health data visualization
I was selected for Communicating Neuroscience; I built these as Excel visuals and tested them on non-expert readers
I built mental-health charts that use color and simpler forms so prevalence and disparity are readable at a glance. I tested the designs with people outside the course, then revised what they could not parse. The rest of the set is in the PDFs.
Major depressive episode prevalence by race, 2005–2021ADHD diagnosis rates across racial and ethnic groups12-month social anxiety prevalence, women vs men
Models & arguments
I write an argument, or a proposal, then report what the result actually said.
John Locke · Commendation
Should we be polite to ChatGPT?
I wrote this for the John Locke Institute Essay Competition, Science & Technology, 2026
I argued that politeness to chatbots is not empty courtesy: it can improve answers, keep civility in practice, and hedge against losing control. I used prompt-politeness studies and moral-psychology accounts of digital manners. The essay received a Commendation in Science & Technology.
I led a proposal for WhiskerWatch, an AI, wearable, and mobile system for earlier pet-health detection. I ran market research on aging owners and urban pets across G20 countries and sketched cross-border pilots. We placed 5th of about 120 teams (Global Bronze).