Projects

Built as a trial project for a local gym franchise, using a public e-commerce churn dataset as a stand-in for their member data. Subscription churn mechanics transfer directly: recurring revenue, cancellation as the target event, and promotional elasticity all behave the same way. The work led to a paid engagement scheduled for Winter 2027.

Developed an end-to-end churn analytics and price elasticity framework evaluating $1M+ in at-risk Customer Lifetime Value (LTV) across nearly 18,000 e-commerce transaction records. Benchmarked baseline Logistic Regression against an ensemble Random Forest Classifier (0.922 ROC-AUC), flagging 76.2% of at-risk churners with 94.2% precision prior to account cancellation.

Uncovered root causes through correlation analysis identifying customer support calls (12.8% importance, +0.288 correlation) and checkout cart abandonment (8.5% importance, +0.251 correlation) as primary churn drivers. Formulated a MECE-based strategy modeling three promotional scenarios (5%, 10%, 20%), finding that a targeted 10% promotion allowed for maximized profits, protecting $50K in revenue (~50% ROI) while preventing margin loss.

Developed a comprehensive financial model for a small-medium enterprise ($40.8M+ total revenue), analyzing its 10-year historical performance (2015–2025, including 2026 financial projections) and forecasting a 5-year trajectory across three macroeconomic scenarios (Bull, Base, Bear). Leveraged Excel for initial data structuring and profit calculations, SQL for KPI aggregation (CAGR, margin trends), and Tableau Public for interactive visualization.

Ultimately provided data-driven M&A recommendations, advising the client to pivot from top-line expansion to disciplined cost control. I recommended a three-year margin improvement phase prior to pursuing an exit to improve business financial health and present a more attractive, profitable asset to potential acquirers.

TaintMonkey is a dynamic taint analysis library for Python Flask apps that uses monkey patching to instrument endpoints without changing source code. It includes a built-in fuzzer and plugin workflow to generate randomized inputs and detect vulnerabilities when tainted data reaches dangerous sinks without proper sanitization. The repo also ships with JungleGym, a dataset of 100+ intentionally vulnerable Flask apps mapped to CWE categories for testing and experiments.

As a researcher, I programmed 32 of the 118 JungleGym test applications and built the corresponding vulnerability plugins. I added JSON-input tainting support to TaintMonkey, fixed major GitHub issues affecting data-flow tracking, and contributed to debugging and core architectural decisions. I also wrote the Experimental Procedures section of our research paper, designed all research figures in Figma, formatted the manuscript in Overleaf, and created the 13-slide deck presented at MIT Undergraduate Research Technology Conference (URTC) 2025. Throughout the project, I coordinated deadlines, kept the team on pace, and actively incorporated feedback to improve the research.