My golf simulator now hosts local LLMs (yatharth.xyz)

A tech enthusiast transformed a personal project into a multi-machine infrastructure by repurposing an Asus NUC and leveraging existing technology to host local large language models (LLMs). Initially seeking a powerful NVIDIA DGX and unable to justify the expense, he turned to cloud solutions but quickly realized the costs were prohibitive for side projects that didn't yet exist. Instead, he utilized a group of unused computers, integrating tools such as Tailscale for stable access, zellij for session management, and automated update services, thus creating a robust environment for ongoing development. Ultimately, he innovatively linked a golf simulator's GPUs to run local models, hilariously highlighting how it became his most efficiently used machine. This development is significant for the AI/ML community as it showcases the feasible transformation of personal computing resources into a mini-datacenter, fostering exploration in AI without high costs. By implementing monitoring and automation with minimal overhead, he demonstrated how technologies like Postgres can efficiently manage and automate workflows. His success underscores the potential for creativity in utilizing existing hardware, paired with AI coding agents, to streamline infrastructure without deep prior expertise, allowing enthusiasts to explore complex projects that might otherwise feel daunting.