AI that lives inside the hardware.

A wearable surface electromyography system that decodes intended movement. The target application is a myoelectric controller: acquire signals from the residual muscles of a partial-hand or transradial amputee, classify the intended action, and actuate a prosthetic hand.

Surface EMG is difficult to acquire. The signal occupies the microvolt range beneath mains interference, motion artifact, and electrode drift.

An FPGA performs acquisition timing and onset detection at deterministic latency and timestamps each muscle event. Embedded cores run the grasp classifier. Grounding topology dominates the design effort; at these amplitudes the fault is almost always the interface rather than the sensor.

Industrial head protection has addressed a single failure mode for a century: impact. It does not record, infer, or reconstruct an incident after the fact.

We are not manufacturing headgear. We develop the AI core and sensor stack that integrates into existing shells, so one architecture serves construction, oil and gas, warehouse, and sport variants. Inference runs on-device, so detection is independent of network availability, and events are written to a timestamped log that neither party can alter.