Techniques for automatically partitioning a multi-platform data transform flow graph to one or more target output platforms are provided. The techniques include performing type inference on a transform graph, wherein the transform graph comprises one or more data transforms, automatically partitioning the transform graph to one or more target output platforms based on one or more policies, performing an optimization of the partitioned transform graph, and generating code, from the partitioned transform graph, for each set of the one or more data transforms based on the one or more target output platforms.