Developed and integrated the HCCL Checkpoint Engine Backend for the volcengine/verl repository, enabling robust support for Huawei Ascend NPU in distributed machine learning training. This work focused on enhancing weight synchronization across distributed nodes by aligning the new backend with the existing checkpoint engine abstraction. Using Python, PyTorch, and Ray, the implementation improved deployment readiness and scalability for Ascend-based environments. The developer also updated documentation and ensured CI compliance, streamlining onboarding and maintenance. By scaffolding future features such as Mooncake transfer engine and Kimi checkpoint integration, the work laid a solid foundation for extensible distributed training pipelines.