Developed and delivered the mHC Post Kernel Enhancement for the ROCm/aiter repository, focusing on optimizing post-stream and res-stream mixing through manifold-constrained Hyper Connection configurations. The work involved implementing new Triton kernels and block configurations, as well as creating comprehensive tests and a benchmark script to evaluate performance. Python and CUDA were used extensively to tune configurations for the gfx950 architecture, while a pytest suite was integrated to ensure correctness and reliability. This contribution established a reproducible benchmark suite and strengthened Triton’s integration with ROCm, providing a robust foundation for improved machine learning workloads within the aiter project.