During October 2025, Actink contributed a targeted performance optimization to the ROCm/rocm-libraries repository, focusing on the triangular packed matrix-vector multiply (tpmv) kernel in rocBLAS. By offsetting the address calculation for the current index rather than recomputing it within kernel loops, Actink reduced redundant arithmetic across both upper and lower triangular configurations, as well as conjugate and non-conjugate operations. This C++-based change leveraged expertise in GPU computing and linear algebra libraries to lower per-iteration arithmetic, improving kernel throughput. The work demonstrated a thoughtful approach to performance tuning, directly benefiting downstream triangular solve workloads without introducing new bug fixes.