The long view, by date.
Every entry the homepage Bento draws from — projects, roles, races, coursework, milestones — sorted newest to oldest.
Data Scientist at Extend
Pricing and risk models for extended warranties, deployed across Peloton, Oura Ring, Sonos, Michaels, and Advance Auto Parts. $600K+ in incremental net revenue from pricing and term optimizations. Behavioral-economics-shaped test design; Bayesian, frequentist, and quasi-experimental evaluation.
Private Equity Analyst at Carrick Capital Partners
Covered cybersecurity and GRC. Built an internal tool that ranks investment prospects from market maps; analysts used it to prioritize sourcing and it set up three meetings at the RSA Conference. Cut analysis time by roughly 80%.
MPS Data Science, Cornell Bowers CIS — graduated, 3.99 GPA
One-year Master’s after the BS. The interesting work was AmEx ML research and the financial-crisis risk project; the rest was the rigor that lets you read a paper without bluffing.
Re-pricing two bank stocks through the 2008 crisis
Rolling-window CAPM in R: alpha, beta, and statistical tests of market-relative performance for Goldman Sachs and JPMorgan through the worst of 2008. The fun was watching beta of an investment bank go from "diversifiable" to "the entire market is the bank" in three months.
AmEx — label-noise mitigation for transformer support-ticket routing
A research-literature-driven pipeline that combined HDBSCAN, Louvain community detection, and an LSTM to find and correct mislabeled training data in a multiclass transformer. Reduced high-confidence misclassifications in production routing; I led the eight-person team.
BS Information Science, Cornell CALS — magna cum laude, 3.88 GPA
Information Science with a Business minor. Independent research in text mining; a lot of late nights in Gates Hall.
Text-mining the American film canon — 359 movie subtitle files
For Text Mining History and Literature, I scraped and processed 359 movie .srt files and ran clustering + sentiment analysis on the corpus. Most of the project was cleaning subtitle files into something the analysis could actually trust.
BERT-based date prediction on a corpus of Morning Brew newsletters
Scraped two years of Morning Brew, hand-rolled the dataset, and fine-tuned BERT to predict newsletter date from sentiment and topical features. Mostly an excuse to learn what fine-tuning actually feels like at small scale.
Private Equity Intern at Carrick Capital Partners
First exposure to software investing. Financial modeling, market research, expert calls. Got asked back full-time the following year.
PwC — operational strategy for CareerVillage
Helped CareerVillage improve student-question to professional-response matching efficiency. Side quest: led a team to 2nd place out of 70+ in PwC’s firmwide consulting case competition, presenting to a Partner panel.