This talk describes a four-step process for constructing a regime-aware portfolio modeling system. The first step is to discover patterns in historical performance in a training and validation period by means of a novel clustering jump model. The objective is to label time periods with relatively homogeneous behavior – called regimes. Next, a forecasting exercise strives to estimate the regimes for the subsequent time-period via random forest and related algorithms. The third step constructs a regime-aware model based on the probability estimates. The final step is to evaluate the process with out-of-sample tests. The approach is interpretable, can be implemented in an asset-only environment or a goal driven framework, and generalizes to a variety of settings, from real-time trading to investors who are seeking to achieve long-term goals. An empirical test shows the advantages of the process.