Developed and integrated the Divergence Proximal Policy Optimization (DPPO) algorithm into the volcengine/verl repository, focusing on reinforcement learning for large language models. The work replaced heuristic ratio clipping with principled divergence-based constraints, such as Total Variation and KL divergence, to enhance training stability and performance. Implementation closely followed the DPPO approach described in recent literature and aligned with the Stable-RL base, with empirical validation on the Qwen3-30B-A3B-Base model using the DAPO dataset. Contributed to robust engineering practices by updating documentation, tagging modules appropriately, and adding comprehensive unit and end-to-end tests to support CI and deployment workflows.