Basically it is a matching eq, isn't it?
Taken from GS forum from one of the guys behind the project,
It is not modelling, in the sense that most people use the term. There is no model that is imposed on the pickup signal. Rather, during training, the algorithm learns all the body resonances that the mic hears, which manifest as both frequency and time phenomena, and that the pickup does not have access to from its vantage point. ToneDexter builds a transform filter, which we call a WaveMap, and applies it to the pickup signal, and this restores almost all of what the mic heard during training.
Could you do this with an EQ? Not practically, due to the 30 to 100 or more adjustment in both magnitude and phase that would have to be made to merely approximate that sound of your guitar's body.