MIT Hackathon Puzzle That Turned Into a Data Science Project

Short summary

A detailed write-up of solving a HackMIT face-customization puzzle by reverse-engineering a hidden scoring API. The author progressed from manual slider tweaking to scripted simulated annealing, coordinate descent, and systematic subset sweeps over 10,000 API calls. Confidence climbed from 60% to 98% before stalling, illustrating how combinatorial optimization beats manual exploration for multi-feature interaction problems.

• •Reverse-engineered a hidden scoring API from a HackMIT avatar puzzle using Chrome DevTools
• •Applied simulated annealing, coordinate descent, and random restarts to search the configuration space
• •Progressed from 60% to 98% confidence but stalled, showing limits of optimization on hidden models

Generated with AI, which can make mistakes.