A mix of sports models, developer tools, and research projects that were interesting enough to keep building long after the original excuse ran out.
A live table that re-ranks Premier League teams by expected points instead of the raw standings.
Built to answer a Chelsea-fan question: were the performances actually as bad as the table made them look? It pairs expected goals with a simpler presentation of the underlying model.
An Elo-based ratings model for tennis matches using historical data stretching back to 1980.
Part prediction engine, part historical visualization. The interesting part was not only rating current players, but making era-to-era comparisons legible.