Worked on the FlagOpen/FlagGems repository to deliver enhancements for tensor operations targeting the GCU300 on the Enflame backend. Focused on backend development and GPU programming, the work involved adding new tensor operation functions and improving error handling to increase compatibility and efficiency in tensor computations. Leveraging Python, the developer addressed backend operations-related bugs, which improved platform stability and reliability. The approach emphasized code quality and traceability, with precise commit messages and clear issue linkage to streamline future maintenance. These contributions enabled higher throughput and broader workload support, advancing the product roadmap and improving the overall developer experience and maintainability.