A novel fine-tuning and evaluation methodology for large language models on IoT raw data summaries (LLM-RawDMeth): A joint perspective in diabetes care

Abstract

This study addresses the challenge of interpreting complex continuous glucose monitoring data in diabetes management by proposing a domain-guided fine-tuning methodology for Large Language Models. Using expert-modeled fuzzy logic datasets and task-aware prompt engineering, the approach enables LLMs to generate accurate, concise, and clinically meaningful summaries from raw glucose data. Experimental results show that fine-tuned GPT-4o achieves superior performance, demonstrating the potential of expert-aligned language models to support medical decision-making and reduce the burden on healthcare systems.