Large Language Model-Assisted Preparation of Transportation Management Plans: A Case Study with WisDOT WisTMP System

A new arXiv paper proposes an LLM-assisted framework that fine-tunes multiple open-source large language models, deployed locally for data security, to automate Transportation Management Plan content generation for the Wisconsin Department of Transportation's WisTMP system. The researchers built a domain-specific dataset by converting historical WisTMP PDFs into structured JSON question-answer pairs, and found fine-tuning significantly improved standard text generation metrics, though the models over-generate strategies and struggle with project-specific justifications and accurate cost estimates. Scaling from 7B/8B to 14B parameters yielded limited gains, and the source code and demo videos will be released at https://zihaosheng.github.io/TMP-LLM/.

arXiv:2610.10650v1 Announce Type: new Abstract: Work zones are critical yet hazardous components of transportation infrastructure, requiring carefully designed Transportation Management Plans TMPs to ensure safety and mobility. However, TMP preparation remains labor-intensive and heavily dependent on practitioner expertise. This paper proposes a Large Language Model LLM -assisted framework to automate TMP content generation, leveraging the WisDOT WisTMP system as the application context. The framework fine-tunes multiple open-source LLMs across different model scales and deploys them locally to ensure data security. To support model training, we construct a domain-specific dataset from historical WisTMP documents by converting PDF files into structured question-answer pairs in JSON format. Experimental results show that fine-tuning significantly improves performance across standard text generation metrics. Further section-wise and strategy-level analyses reveal that, while LLMs achieve strong overall performance, they tend to over-generate strategies and struggle to produce project-specific justifications and accurate cost estimates. In addition, scaling from 7B/8B to 14B yields limited gains. These findings demonstrate the potential of LLMs to improve TMP preparation efficiency while highlighting remaining challenges in LLM-assisted TMP development. The source code and demo videos will be publicly available at https://zihaosheng.github.io/TMP-LLM/.