Rillomas developed targeted GPU performance enhancements for matrix workloads in both the ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp repositories. Focusing on Intel Xe2 GPUs, Rillomas implemented architecture and runtime detection in C++ to enable the VK_KHR_cooperative_matrix Vulkan extension exclusively on supported hardware. This approach ensured that performance gains from cooperative matrix operations were delivered to Xe2-class devices while maintaining compatibility and stability for older GPUs. The work demonstrated a strong grasp of driver development, GPU programming, and performance optimization, with careful feature gating to minimize regression risk and ensure robust, hardware-aware improvements across multiple codebases within a single month.