Fitness social network that actually kept people moving — built the backend and ML recommendation engine that matched users with workouts, trainers, and accountability partners. The founder had a vision for AI-driven fitness personalization but the existing codebase was a monolithic Nest.js app that couldn't handle real-time interactions. I introduced Nx monorepo for shared code, Redis pub/sub for live workout sessions, and a collaborative filtering ML pipeline that boosted engagement 3x.