A Study of LLM-based Conversion from PDF to Markdown for Local Government Documents

Abstract

In recent years, the digitization of public administrative documents has progressed. However, many municipal documents are still published in PDF format, which poses significant challenges to effective data integration and reuse. This study aims to improve interoperability by converting such unstructured PDF documents into structured formats using multiple large language models (LLMs). We conducted a comparative evaluation of five LLMs, to measure their ability to extract and restructure information from municipal PDFs into Markdown format. These results show that LLAMA-4-maverick demonstrated the highest accuracy in identifying both contextual and structural features without misinterpretation. These findings suggest that advanced LLMs can be effectively applied to the task of document structuring.