Rillomas developed targeted GPU performance enhancements for matrix workloads in ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp by enabling the VK_KHR_cooperative_matrix extension specifically for Intel Xe2 GPUs. Using C++ and Vulkan API, Rillomas implemented both architecture and runtime detection to ensure cooperative matrix features were only activated on compatible Xe2-class hardware, maintaining backward compatibility and minimizing regression risk for older GPUs. This approach allowed both repositories to leverage hardware-aware optimizations without compromising stability. The work demonstrated depth in driver development and performance optimization, delivering focused improvements for GPU-accelerated applications while carefully managing feature rollout across multiple codebases.