Simon Parkinson

This paper demonstrates that fine-tuning small language models (SLMs) on a synthetic, solution-rich Windows event log dataset allows them to outperform larger LLMs in identifying issues and providing actionable remediation while maintaining low computational overhead.