An LLMs-based Multi-Agent Collaborative Framework for Intelligent Public Fund Allocation

With the continuous improvement of e-government platforms and the rapid development of artificial intelligence technology, an innovative public fund allocation model has emerged, which achieves precise allocation of public funds through automatic filling of application forms and intelligent verification of enterprise qualifications. However, existing studies still lacks systematic modeling of the implementation mechanism of this public fund allocation model. In response to the inherent complexity of the operating logic of this model and the unstructured characteristics of policy texts and enterprise data, this paper innovatively proposes a large language models-based multi-agent collaborative intelligent allocation method for public funds. This method consists of three core intelligent agents: policy agent, enterprise agent, and match agent. Among them, the policy agent is responsible for parsing policy texts and simulating government decision-making processes. The enterprise agent constructs the enterprise profile by integrating structured and unstructured data, and automatically generate funding application forms. The match agent is responsible for outputting the matching results and their decision basis. Through experimental research on a real-world dataset, this paper verifies the effectiveness of the proposed method. © 2025 held by the owner/author(s).