Interested in the intersection of technology and business

Esfan Daya

About Esfan

I’m a student at the Wharton School of the University of Pennsylvania, studying Statistics, Data Science, and Finance. I spend most of my time exploring how AI and machine learning can make business systems smarter and more efficient — from automating private equity workflows at Woodvale to building an ML startup that helps pharmacies forecast drug prices.

When I’m not building models or chasing new ideas, I’m usually chasing something else — a finish line. I’ve been running competitively for most of my life, from setting high-school records and earning All-American honors to racing Division I for Penn. Running has taught me focus, discipline, and how to stay calm when the world blurs by at 4:10 per mile. It’s also given me the chance to give back — through the Young Quakers Program, I mentor students from West Philadelphia who share that same spark for running, helping them build confidence and community through the sport.

Off the track, I trade spikes for strategy — I’ve been an avid chess-tournament regular since childhood and a casual poker enthusiast ever since I discovered the joy (and heartbreak) of a well-timed bluff. And when it’s time to unwind, you’ll probably find me behind a mic, singing for a small band with my roommates.

Projects

  • https://github.com/esfan/drug-price-neural-prophet-analysis

    -Founded an ML-driven startup that optimizes drug purchase timing for independent pharmacies; built forecasting models (NeuralProphet, XGBoost, etc.) with macro indicators and trading-style execution (dynamic scaling, z-score, confidence-weighted optimization), achieving ~10% cost savings, reducing waste, and preventing any drug shortages in a deployed MVP

    -Engineered end-to-end quantitative pipeline: scraped and processed 10+ years of Medicaid and McKesson data, built automated feature selection to prevent leakage, and implemented correlation and lag analysis across 20+ economic variables

    -Secured pre-seed investor commitments and built a network of partner pharmacies for broader rollout post-MVP

  • https://github.com/esfan/pe-deal-flow-automation

    -Engineered AI-driven automation tools (custom email generation, property sourcing and data scraping, deal flow tracker integration) that streamlined underwriting and fundraising processes across $100M Woodvale Opportunity Fund II

  • -Developed a data-driven framework to analyze multi-year wheat yield and pricing trends across global markets. Integrated historical supply–demand datasets, weather patterns, and macroeconomic indicators into Python-based models to identify anomalies and predict yield outcomes. Built regression and machine learning models to forecast production and price fluctuations, supporting trading strategy design.

    -Engineered quantitative back-testing systems that simulated commodity trading strategies using technical indicators such as moving averages, MACD, RSI, and Bollinger Bands. The models incorporated stock–commodity correlation analysis, EBITDA forecasting, and options-based hedging under varying geopolitical scenarios, enabling the team to assess strategy resilience and profitability under real-world conditions.

  • -Developed an advanced machine learning pipeline to predict and explain default risk across a large portfolio of small business loans. Combined structured financial data with borrower demographics and macroeconomic indicators to build feature-rich datasets, applying techniques such as outlier filtering, multicollinearity reduction, and principal component analysis for dimensionality control.

    -Trained and compared multiple classification models—including logistic regression, random forest, XGBoost, and ensemble stacking—optimizing hyperparameters through cross-validation to maximize AUC and F1 scores. Conducted SHAP value analysis to quantify variable importance, uncovering key drivers of default such as leverage ratios, credit aging, and revenue volatility.

    -Presented findings through an interactive dashboard and narrated analysis video, demonstrating how model insights could be operationalized to improve bank risk scoring systems, portfolio stress testing, and credit policy decisions.