I see nothing to preclude a foundation model being augmented by a smaller model that serializes particulars about an individuals cumulative interaction with the model and then streamlines it into the execution thread of the foundation model.