Personalized alternative medicine holds substantial promise for holistic healthcare; however, systematic optimization of multi-herb, multi-target interventions is still an open problem in terms of computational difficulties associated with combinatorics, nonlinearity, and individual differences. This paper develops a comprehensive mathematical optimization approach to personalized alternative medicine interventions by combining three approaches: evolutionary algorithms to optimize prescription of herbs, reinforcement learning to adapt the therapy, and Bayesian multidimensional hierarchical models to characterize patients’ responses to the medication. The effectiveness of the proposed optimization framework is validated through experimental analysis utilizing clinical records from traditional Chinese medicine (n=5,216). It is found that the optimized prescriptions with the use of evolutionary algorithms result in 28.5% higher effectiveness than the conventional methods (95% CI: 18.7-37.3%).