Jiapli worked on enhancing device management and cross-backend interoperability in the ROCm/pytorch and pytorch/pytorch repositories. Over two months, they implemented PrivateUse1 device type support by refactoring the parse_type function, enabling broader device compatibility and improving code maintainability. In subsequent work, Jiapli developed a shallow copy compatibility operator to allow safe tensor copying between CPU and PrivateUse1 backends, updating core TensorImpl and dispatch-key logic to support future extensibility. Their contributions, primarily in C++ and Python, focused on backend development, device management, and tensor operations, demonstrating a thoughtful approach to maintainable, extensible code and addressing evolving interoperability requirements.