This study presents TRACK, an AI-driven Intelligent Tutoring System (ITS) for personalized mathematics learning, grounded in the Technological Pedagogical Content Knowledge (TPACK) framework. TRACK leverages reinforcement learning and Natural Language Processing (NLP) to deliver adaptive difficulty scaling, context-aware feedback, and interactive simulations. A simulation with 300 virtual students shows TRACK outperforms state-of-the-art systems (ALEKS, GeoGebra TUTOR, Cognitive Tutor), achieving a 21.4% higher retention rate (85.0% vs. 70.0%), 33.3% faster feedback (0.8 vs. 1.2 s), and 12.5% greater knowledge gain (18.0 vs. 15.0 points). Despite challenges like scalability and limited content scope, TRACK’s TPACK-driven design highlights AI’s potential to transform mathematics education, with future directions including real-world validation and broader curriculum coverage.