This document summarizes the evolution of the algorithms behind Deezer Flow, Deezer's personalized music streaming service. It began with an initial system (2014) that generated tracklists offline based on user likes and recommended albums. A content tagging system (2015) was developed to better understand tracks and build user profiles. This led to live adaptive algorithms that analyze user behavior and query an index to generate more personalized, real-time tracklists based on tags, popularity, and user feedback. The system continues to advance towards fitting music to user mood and context while improving performance for new users.