A homogeneous hidden Markov model has recently been utilized in the diagnosis of daily monsoon rainfall over a small network of stations in central-western India. A four-state decomposition inferred using this model was found to capture large-scale flow regimes corresponding sensibly with recognized features and behavior of the monsoon, including breaks and active phases, the intraseasonal oscillation and monsoon-ENSO interaction. The analysis also provided evidence for a linkage between intraseasonal and interdecadal monsoon variability. Here, this work is extended into the predictive domain via the deployment of a nonhomogeneous model, in which the relation between modeled states and the large-scale flow is inverted, and the latter permitted to modulate the Markov transition probabilities. Screening of candidate climate models, selection of appropriate circulation indices, statistical model fitting and validation and coupling with both observational and modeled circulation fields will be discussed, and potential predictive skill assessed.