Offline Edge-AI Security System 

I built an offline AI security and access-monitoring system that runs entirely on local hardware without relying on Wi-Fi or external cloud services. The system uses YOLOv8 for real-time person detection and tracking, then conditionally activates facial recognition only when a person is close enough for reliable identification. This reduces unnecessary inference, improves efficiency, and keeps the system focused on high-confidence recognition events.

The pipeline combines object detection, proximity-based triggering, face embedding comparison, local identity storage, and automated alerting. Known individuals are represented using facial embeddings stored in JSON files, allowing the system to quickly compare incoming face data against registered profiles. When an unrecognized person is detected, the system sends an automated text alert to the property owner.

This project demonstrates my ability to build practical computer vision systems that balance accuracy, latency, privacy, and deployment constraints. By keeping all inference and identity matching on-device, the system avoids unnecessary network dependence while maintaining fast, reliable monitoring in real-world environments.