Portfolio
Sports research, applied modeling, algorithms, and data systems. The larger projects are documented as complete systems rather than isolated notebook exercises.
A destination-specific extension of projected WAR that estimates a player's marginal effect on expected championships and playoff appearances over a competitive window. The frozen framework combines player forecasting, team-state decay, structural franchise strength, correlated simulation, paired counterfactuals, and a fixed league reference.
A production-oriented forecasting project that generates projected scores, win probabilities, model point spreads, and totals for NFL and FBS games. Version 0.1 uses dynamic Elo, recent scoring form, rest, and home-field context with time-ordered validation; richer play-level and roster features are being added only when they improve walk-forward performance.
A production platform for a student-run league that combines public statistics and history with the operational tools needed to run a season. The public interface covers standings, schedules, player and team pages, sortable leaderboards, awards, playoff projections, power rankings, and an interactive historical grid. Behind the public site are stat-ingestion workflows, database migrations, commissioner tools, authenticated captain scheduling, and validation logic for series and player data.
A five-class geolocation image-classification project built in PyTorch. The workflow processed 49,995 street-level images from 122 countries, center-cropped and resized them to 224×224, and used an ImageNet-pretrained ResNet18 with a modified classification head. Weighted sampling and inverse-frequency cross-entropy addressed class imbalance. The final holdout evaluation reached 68.1% accuracy across 9,999 test images.
A collection of Python implementations developed in Topics in Applied Mathematics. The work includes bucket sort with insertion sorting, randomized quickselect, maximum-probability path estimation for Markov chains using log probabilities, and graph search that excludes nodes within a threshold distance of hazards before computing a shortest feasible path with NetworkX and Dijkstra's algorithm.
Course and research analyses covering exploratory and confirmatory factor analysis, structural equation modeling, generalized additive models, reliability analysis, modification indices, spline interpretation, and reproducible reporting in R and Python.