Unlocking Large Language Models for Project Scheduling

The timely delivery of the project is a significant concern for project stakeholders. Effective project scheduling is essential to ensure that projects are delivered on time. However, project scheduling requires expertise in complex decision-making. Especially when considering resource factors, project scheduling becomes a complex mathematical optimization problem. This professional barrier hinders the implementation of project scheduling. To address this problem, this study proposes an approach of Large Language Models (LLMs) as agents for project scheduling (LAPS) to assist with project scheduling by solving two problems: Critical Path Problem (CPP) and Resource-constrained Project Scheduling Problem (RCPSP). The LAPS approach has three parts: (1) Code agent: this part converts project information expressed in natural language into a programming language for subsequent calculations. (2) Scheduling agent: for the CPP, this agent calculates scheduling results directly. For the RCPSP, this agent invokes optimization models to generate scheduling results. (3) Feedback agent: this part converts the scheduling results expressed in programming languages into natural languages and provides feedback to project stakeholders. The effectiveness of the LAPS approach is tested in comparative experiments. The results demonstrate that the LAPS approach can significantly improve the efficiency of project stakeholders in completing project scheduling. A maximum time saving of 49.2% can be achieved for stakeholders with expertise. It narrows the time gap between project stakeholders with expertise and those without expertise to 7.9%. This study contributes to the rapidly expanding field of LLMs-based intelligent project management.