Scalable Invoice Reconciliation for SMEs: Edge-Deployed LLMs in Multi-Agent Systems

Large Language Models (LLMs) have transformed business operations through their advanced capabilities in logic processing, contextual understanding, and task execution. However, deploying multi-agent workflows in real-world scenarios presents significant challenges, including high computational demands, elevated costs, data privacy concerns, and the lack of mature frameworks. This paper introduces a novel approach to invoice reconciliation utilizing edge-deployed LLMs within a multi-agent system specifically designed for small and medium-sized enterprises (SMEs). Our solution addresses critical challenges such as resource efficiency, data privacy, and cost-effectiveness while delivering high accuracy. Through a comprehensive evaluation, we demonstrate that the mid-tier model qwen2.5:32b achieves an impressive 99.9% success rate with execution times comparable to proprietary models. Additionally, the functionary-small model achieves an optimal balance between success rate and resource consumption, making it particularly suitable for resource-constrained, real-time reconciliation tasks. By leveraging localized LLM deployments and efficient agent coordination, our system exhibits scalability, adaptability, and practical utility in automating invoice reconciliation processes for SMEs.