Akeel Ather MedinaMSc High Performance Computing, University of Edinburgh
I’m broadly interested in the systems side of machine learning and HPC. See more here.
I recently finished an MSc in High Performance Computing at the University of Edinburgh, where my thesis profiled the time and energy cost of KV-cache offloading during LLM inference on H200 (PCIe) and GH200 (NVLink-C2C) systems. With TeamEPCC at the ISC26 Student Cluster Competition, I optimized llama.cpp inference under a power cap and built a BeeGFS parallel file system across a heterogeneous cluster. Before that, I spent two years at Abyss Solutions, an Australian startup, building point-cloud analysis pipelines at scale.
I’m currently looking for PhD positions and engineering roles in GPU systems and LLM inference.
Short paper reviews.
Yan, Karlin, and Grant · arXiv 2026
Reverse engineering the NVIDIA driver command streams.
Ham et al. · MICRO 2024
M²NDP puts a general-purpose RISC-V vector engine inside the CXL memory controller, builds the kernel launch path out of ordinary CXL.mem loads and stores.
Articles and reflections.
A musical side project.
Oct 15, 2026 · 14 min readAn ode to research, and what I learned through my thesis.
Sep 11, 2026 · 3 min readTuning llama.cpp, building a parallel filesystem under competition deadlines, and what the SCC actually tests.
Sep 10, 2026 · 5 min read