Have you looked into Mutable Grids? It's the same principle -> scanning machine learned rhythmical tables for kick, snare and hihat + individual accent outputs. There's also a similiar rhythm generator on MI Marbles red T-mode with less outputs, and also an awesome stochastic (basically ML with probabilities) model hidden away in the code base (needs modified firmware to access).
For deep euclidean sequencing with great CV-control + lots of outputs Pamela's New/Pro Workouts are GOAT material.