Sort the rock —
in real time, on the edge.
Vision models deployed across a fleet of sorters, running on the machine — each one reads the rock as it flows and acts in real time to split value from waste. Models, fleet and the assay data that trains them, on one sovereign platform.
Value or waste — decided too late, by hand.
On a sorting line, the difference between valuable rock and waste is decided in a fraction of a second — yet the intelligence to make that call sat in a lab. Models lived on a data scientist's laptop; sorters ran on fixed rules; assay results came back weeks later to tell you what you'd already let through. And in a regulated jurisdiction, none of that data could leave the country.
What the system handles.
Models, deployed to the whole fleet
Computer-vision models are packaged and pushed to every sorter, running on the machine's own GPU. Retrain once, update the model, and the entire fleet sorts to the new spec — no truck rolls, no reflashing by hand.
Real-time sorting at the machine
Each sorter reads the rock as it flows and the platform drives its actuators in real time — value diverted from waste in milliseconds, where the rock is, not in a control room a network hop away.
The fleet, operated as one
Every sorter monitored live: throughput, model version, health, what it's diverting and why. A line of machines run as a single fleet, with a full trace on every decision.
Assay-grounded & sovereign
Lab chemistry and field analysis feed back to sharpen the models, closing the loop from rock to result. The whole platform runs on sovereign hosting in your own jurisdiction — open-source, auditable, no sample or sensor data crossing a border it shouldn't.