Worked on enhancing the ROCm/aiter repository by developing new configuration files for gfx942 Triton A8W8 blockscale GEMM operations, specifically targeting GLM-5 shape workloads with diverse input and output sizes. Leveraged configuration management and GPU programming skills to centralize and streamline GEMM configurations, reducing the need for manual tuning and improving deployment readiness. Utilized JSON to define flexible parameters, enabling faster onboarding of GLM-5 workloads and supporting future performance optimization. Collaborated through co-authored pull requests, ensuring traceability and maintainability of changes. The work positioned ROCm/aiter for broader hardware support and facilitated more efficient performance tuning for complex GPU workloads.