Worked on a comprehensive rewrite of the NLMeans Vulkan filter in the FFmpeg/FFmpeg repository, focusing on both performance and quality improvements for video denoising. The approach involved increasing parallelism, splitting the weights pass into vertical and horizontal components, and removing the atomic float requirement to streamline GPU processing. Addressed visual artifacts and improved stability by fixing image loading and buffer handling within the Vulkan path. Implemented memory bandwidth optimizations to reduce GPU-CPU data transfer overhead. Utilized C and GLSL to deliver these enhancements, demonstrating expertise in GPU computing, performance optimization, and video filtering for high-quality media processing pipelines.