Here the authors develop an AI framework called E-MAF that, starting from a desired release profile, couples process-aware prediction with expert-mimic evolutionary search to generate interpretable, lab-ready formulations for controlled release systems.