Worked on enhancing the numerical robustness of the matrix multiplication (mm) kernel in the FlagOpen/FlagGems repository. Focused on addressing indexing correctness and overflow risks by implementing int64-based index calculations, which improved compatibility and stability for larger matrices and multithreaded workloads. Utilized Python and leveraged expertise in GPU programming, matrix operations, and parallel programming to resolve a critical bug affecting kernel reliability. The changes ensured safe operation under high concurrency and laid the groundwork for future scalability. This work contributed to more reliable matrix computations, particularly in environments where large-scale data and parallel execution are essential for performance and accuracy.