BoostAPR is a three-stage framework for automated program repair that uses reinforcement learning with dual reward models to improve edit-level feedback. It achieves competitive results on various benchmarks, particularly in cross-language generalization.