Solo, week-long hackathon in the University of Texas at Austin – McCombs post-graduate program in AI & Machine Learning: Business Applications (September 2026), separate from the program's four graded builds. Built with Claude Cowork, Anthropic's desktop agent, which wrote and ran the notebook; the brief, the review of the results and the three submissions were my own. A regression model predicting annual revenue for 500 restaurants from 33 columns (location, opening date, cuisine and theme, social-media popularity, ratings and survey scores), trained on 3,493 labeled restaurants. 2.7% of restaurants caused 66% of the squared error, so the training labels were capped at the IQR bound before any tuning; seven models were compared and tuned Gradient Boosting was chosen first (9.4% better than predicting the average on held-out RMSE), then replaced by a tuned CatBoost blend (four models, seed-averaged) chosen by cross-validation. Leaderboard: 9th of 53 on the first submission (₹12.42M RMSE), 7th after retraining on all labeled data (₹12.20M), and a final 6th of 53 (₹12.05M) when the board closed on 28 September 2026. Write-up, notebook and slides: https://averyresume.com/mccombs.html#hackathon