TRACS (Tracking and Recognition through AI-integrated CCTV System) came from the challenges we face in India. Many CCTV systems are passive and do not prevent incidents as they happen. We developed an AI system that identifies unusual behaviors like loitering, tailgating, and pacing using YOLOv8, SORT, and Isolation Forest. It includes a scoring algorithm, zone sensitivity, and cooldown logic. The system also supports forensic tracking through facial recognition, number plate OCR, and appearance-based re-identification. Using Python, OpenCV, Flask, and S3/MinIO storage, we built a modular, edge-compatible prototype. We addressed issues like tuning unsupervised models, keeping real-time speed, and minimizing false positives. We are proud to have created a scalable system that works for both prevention and investigation. Through this, we learned to balance AI accuracy with real-world limits. Next, we plan to add a React dashboard, integrate with CCTNS, and deploy it in a real-world campus setting.