Functional grading has recently seen renewed interest with the advancement of additive manufacturing. Unfortunately, the integrity of functional gradients in alloys tends to be compromised by the presence of brittle phases. Recently, CALPHAD-based tools have been used to generate isothermal phase diagrams that are in turn utilized to plan gradient paths that avoid these phases. However, existing frameworks rely extensively on the (limited) ability of humans to visualize and navigate high-dimensional spaces. To tackle this challenge, a Machine Learning approach was used here to map out undesirable regions as ‘obstacles’, while a path-planning algorithm, commonly used in robotics community, was utilized to identify a path in a composition space that would avoid the obstacles, while simultaneously minimizing a cost function. This framework was validated by designing and 3-D printing a functional gradient in bulk samples from 316L stainless steel to pure chromium with a multi-material direct laser deposition system. Both the planned gradient and simple linear gradient samples were fabricated and characterized in as-deposited and heat-treated states to determine local compositions, microstructure and phase constituents. The planned gradient resulted in complete elimination of the detrimental σ phase after heat treatment, demonstrating the success of the methodology.