TRACK Families is General Motors’ voluntary onboarding program that matches small groups of new hires for casual Detroit events (happy hours, dinners, hikes, etc.) so they actually make friends and feel at home instead of just showing up to work. The original algorithm was a single massive k-means cluster that produced 60% perfect groups, 40% total disasters, and left ~10% of new hires completely unassigned because they lived too far away from everyone else. I replaced it with a three-stage, human-in-the-loop pipeline that first builds probability heat maps from ~150 historical participants to generate realistic synthetic training data for tuning variable weights, then pre-clusters larger cohorts (15–20 people) by commute distance with a modified k-means, and finally forms the actual event groups of 4–6 based on personality and activity preferences with easy coordinator overrides. The new system now achieves 100% assignment, near-perfect turnout, zero complaints, and—most importantly—coordinators actually love using it.