During June and July 2025, Etasnadi developed Vulkan-accelerated convolution operations for the ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp repositories, focusing on both 1D and 2D convolution support. He implemented new shader code and pipeline configurations in C++ and GLSL, enabling efficient GEMM-based execution paths for deep learning inference on GPUs. His work included comprehensive testing, performance validation, and cross-platform optimizations to ensure correctness and hardware compatibility. By integrating these features into the existing deep learning frameworks, Etasnadi improved model performance and scalability, demonstrating strong skills in GPU programming, shader development, and performance optimization within complex codebases.