Haak M1
1XENON Systems, Springvale, Australia
Biography:
Malcom Haak joined XENON Systems in 2025 as a talented Solutions Architect/Engineer, based in Queensland. Malcolm is a seasoned professional, with over 18 years of experience in IT, 13 years of which has been in High-Performance Computing (HPC). He has a strong focus on HPC having worked for SGI, DDN and HPE in the past with a particular interest in storage and AI workflows.
He has successfully contributed to numerous large-scale deployments, including projects with the Bureau of Meteorology, UK Met Office and University of Bristol's Isambard-AI, where his problem-solving abilities truly shine. Malcolm has also worked for the National Computational Centre (NCI) in Australia and is highly regarded. Malcom is passionate about leveraging his expertise to design innovative solutions that address complex challenges, making him a valuable addition to the XENON Systems team.
His career reflects a deep understanding of both specialized HPC domains and general IT practices, allowing him to create innovative solutions that bridge across diverse technical landscapes. He is committed to lifelong learning, driven by a passion for exploring new technologies and delivering results in the ever-changing field of HPC.
Abstract:
Supercomputers are powerful, but they do not automatically make inefficient code fast. For some researchers, programming is a necessary tool rather than their primary expertise, and small issues in scripts, workflows, or job configurations can quietly consume valuable compute quota.
This talk explores how locally hosted LLMs can help researchers write, review, and improve the code they use for high-performance computing. From identifying bottlenecks and debugging batch scripts to improving data handling, parallelization strategies, and reproducibility, local models can act as always-available coding assistants without requiring sensitive research data or code to leave your environment.
We’ll discuss practical ways to use local LLMs to avoid common performance pitfalls, make better use of allocation hours, and build more reliable computational workflows. The goal is not to turn every researcher into a software engineer, but to show how local AI tools can help them get more science from the compute resources they already have.
