Work
Featured Project
Healthcare Price Transparency
Why does the same surgery cost 10× more across the street?
Built an end-to-end ML pipeline on ~300M hospital price records (27 GB) spanning 1,562 hospitals across 44 states. Dask-based cleaning distills the raw data into a 227K-record modeling matrix; Naïve Bayes classifies pricing tiers at 0.87 accuracy (ROC-AUC 0.96), with HDBSCAN clustering and association-rule mining surfacing four distinct hospital market segments.
01
Treasury Yield Regime Forecasting
Regime-aware neural networks (FNN, CNN, LSTM, Transformer) forecasting the 10-Year U.S. Treasury yield from a decade of FRED macro, market, and Fed policy data. A domain-informed hybrid model reaches R² ≈ 0.91 out-of-sample, with Fed-funds and CPI shock simulations.
02
ATP & WTA Tennis Match Predictor
Predicts pro singles winners on the ATP and WTA tours using overall and surface-specific Elo plus XGBoost, trained on match data from 1991 to the present. Beats rank and Elo baselines on 2022+ held-out matches (AUC 0.716 ATP / 0.713 WTA) — and you can try the model live, right in your browser.
03
EPL Transfer Narrative NLP
NLP + network science on Premier League transfer press coverage and Reddit discourse from 2016–2025. Quantifies how hype independently predicts competitive outcomes beyond financial spend alone.
04
MLS Narrative Network Analysis
PageRank on press and Reddit co-occurrence networks for 28 MLS clubs — ~7,200 articles and ~15,800 posts, 2018–2024. Narrative attention tracks performance (r = 0.41) but fails to predict next-season points, a rigorous null result — plus a causal estimate of the Messi signing's media effect.
05
Household Electricity Demand Forecasting
PyTorch LSTM forecasting daily household electricity consumption from multi-year power usage data structured as supervised learning sequences. Evaluated on RMSE, MAE, and R².
06
Telco Churn — TabNet
End-to-end churn analytics on 7,043 IBM Telco customers using TabNet's attentive interpretability, benchmarked against logistic regression, random forest, and XGBoost. Surfaces high-risk segments (42.7% month-to-month churn vs. 2.8% two-year) and frames outputs as ROI-aware retention targeting.
07
California Mall Sales Analysis
Exploratory analysis of 99K+ sales transactions across 10 malls (2021–2023), profiling customer segments and mall characteristics to identify drivers of sales performance and inform marketing strategy.
08
Nvidia Financial & Strategic Analysis
Comprehensive analysis of Nvidia's strategic position — supply chain, GPU/AI market dynamics, competitive moat, and DCF-based financial forecasts with pro forma modeling.
09
OCTA Travel Demand Forecast
GIS and transportation modeling for the Orange County Transportation Authority — analyzing population growth, traffic flow, and bikeway usage to support infrastructure and land-use planning.
