Running YOLO at the pit edge when the uplink drops

At most remote sites, connectivity is the first thing to fail and the last thing to come back. A cloud-only vision system goes blind the moment the link drops, which is exactly when you can least afford it. So the platform was designed offline-first from day one.

The platform is designed so each site can run a ruggedized edge node that ingests existing camera feeds and runs detection locally. Models are quantized to run in real time on modest hardware, so a PPE breach or intrusion can be flagged on-device — with or without a connection to the outside world.

When the uplink is available, the node syncs detections, clips, and telemetry to the cloud digital twin. When it isn't, everything is buffered and forwarded the moment the link returns, so the record stays continuous across outages.

The goal is a system that behaves the same on a clear day in town as it does in a dust storm hundreds of kilometers from the nearest tower. That reliability — not raw model accuracy — is what makes edge AI usable in the field.