Worked on the NVIDIA/cuda-python repository to address memory management issues in DLPack capsule creation. Focused on fixing memory leaks that occurred during interrupted capsule creation by implementing robust cleanup routines, including null-pointer checks and proper deallocation of resources. Enhanced the reliability of the codebase by expanding test coverage to validate cleanup and error-path resource management, ensuring that resource lifecycles are handled correctly even in failure scenarios. Utilized Cython and Python testing frameworks to strengthen defensive programming practices. The work improved the stability of CUDA data structure interoperability and contributed to higher code quality through careful attention to memory management details.