Frontiers Market - Edge AI NVR System

Livestock monitoring in rural areas suffers from poor internet connectivity and prohibitive cloud ingress costs for video streams. Traditional off-the-shelf NVR solutions were too bandwidth-heavy and lacked the specialized AI capabilities needed for cattle management. Video streams would frequently disconnect, and cloud-based processing was both expensive and unreliable in areas with limited connectivity.

Engineered a proprietary Edge AI architecture using Raspberry Pi 5 clusters as local compute nodes. Built custom NVR software that processes RTSP video streams locally. Confirmed architecture: Raspberry Pi clusters did pure NVR (capture/recording), with some Pi units also running experimental on-device YOLO tracking; Nvidia Jetson units handled real edge inference — both weight calculation and YOLO tracking; the cloud ran the larger, fully-trained model for bigger/heavier-compute events where the extra round-trip latency was an acceptable tradeoff for more accuracy. Implemented a "store-and-forward" mechanism that queues critical data and intelligently syncs when connectivity is available, ensuring zero data loss. The system operates autonomously for days without internet access, maintaining full monitoring capabilities.

A low-bandwidth, local-first monitoring system that processes RTSP video on-device, with no published cost or uptime figures. The local-first approach removed the dependency on consistent internet connectivity, and the custom NVR runs computer-vision models at the edge for cattle monitoring that off-the-shelf tools don't handle. It replaced costly third-party NVR subscriptions, cutting that upfront and subscription cost.