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

Authors: Zihao Sheng, Pei Li, Zilin Huang, Yen-Jung Chen, Yuhao Luo, Zhengyang Wan, Steven T. Parker, David A. Noyce, Sikai Chen

Subjects: Computation and Language (cs.CL)

Cite as: arXiv:2610.10650 [cs.CL] — Submitted on 7 Oct 2026

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, the authors 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/

Key Contributions

1. LLM-assisted TMP automation framework — A pipeline for automatically generating TMP content, applied to the WisDOT WisTMP system.
2. Domain-specific dataset — Historical WisTMP documents converted from PDF into structured question-answer pairs in JSON format to support fine-tuning.
3. Multi-scale open-source model evaluation — Multiple open-source LLMs fine-tuned across different parameter scales and deployed locally to preserve data security.
4. Detailed performance analysis — Section-wise and strategy-level evaluations identifying strengths (overall generation quality) and limitations (strategy over-generation, weak project-specific justifications, and imprecise cost estimates).



Key Findings



• Fine-tuning substantially improves performance on standard text generation metrics.
• Scaling model size from 7B/8B to 14B parameters produces only limited gains.
• LLMs tend to over-generate strategies and underperform on project-specific justifications and cost estimation.
• Local deployment of open-source models supports data security requirements in public-sector workflows.



Resources



• arXiv: 2610.10650 [cs.CL]
• Code and demo videos: https://zihaosheng.github.io/TMP-LLM/



via ArXiv CL+LG