Worked on backend development and resource management across two repositories, focusing on reliability and code quality. In the GeeeekExplorer/nano-vllm project, developed a CUDA Graphs Resource Management feature using Python and CUDA programming, ensuring resources were released before program exit to prevent memory leaks and optimize resource usage. This approach stabilized the CUDA Graph lifecycle, supporting production-ready graph execution. Later, contributed to dataease/sqlbot by addressing a critical bug in the chat rename flow, implementing input validation to prevent invalid renames and improve data integrity. Emphasized robust Python development practices, prioritizing safe resource handling and stable user experiences throughout both projects.