Digica

The preventive-maintenance work is documented the way an engineering buyer would want it: a neural network with XGBoost decision trees and SHAP for explainability, predicting mobile-device failure at 91% accuracy.

Sensor-fusion work covers training convolutional networks on synthetic images and combining RGB with near-infrared, and there is published work running language models on Arm CPUs for edge devices—a constraint most AI agencies never encounter.

Clients are named as logos rather than attached to engagements, so the technical evidence is strong and the commercial evidence is not.

Best for manufacturers and device makers who need models running on hardware rather than in a cloud region, and who will recognize the difference immediately.