Traditional forensic investigations rely heavily on passive CCTV monitoring and manual sketch identification, resulting in subjective, time-consuming, and delayed suspect tracking. This paper introduces the Nexus Forensics AI Sys-tem, an end-to-end, multi-tier active surveillance architecture designed to bridge the modality gap between static forensic records and live video feeds. The proposed system orchestrates a three-tier technology stack comprising a React-based interactive frontend, a Node.js/Express backend for data logistics, and a highly optimized Python/FastAPI AI engine. To ensure real-time efficiency across multiple IP and web cameras, the com-puter vision pipeline utilizes a two-stage detection mechanism: Ultralytics YOLOv8s for structural person detection, followed by RetinaFace for precise facial boundary extraction. Biometric matching is executed using the ArcFace model, calculating cosine-distance matrices against stored identities. Upon a confirmed match, the system automates evidence recording, captures live GPS telemetry, and dispatches immediate forensic alerts. This architecture demonstrates a scalable, proactive approach to modern law enforcement and smart city security.

Index Terms—Forensic Surveillance, Face Recognition, Deep Learning, YOLOv8, ArcFace, Real-time Tracking, Biometrics.