Search

A-X-MAS: An Auditable, Explainable Multi-Agent System Design and Implementation

Author(s)

Yutong Yang1, Yuping Wang2,*

Affiliation(s)

1School of Telecommunications Engineering, Xi'an University of Electronic Science and Technology, Xi'an, China 2School of Computer Science and Technology, Xi'an University of Electronic Science and Technology, Xi'an, China *Corresponding Author

Abstract

Recent advancements in artificial intelligence have promoted the application of multi-agent systems in complex decision-making scenarios. However, the adoption of such systems in high-stakes domains including cross-organizational process management is limited by insufficient transparency and auditability. This paper presents the design, implementation, and evaluation of A-X-MAS, an Auditable and Explainable Multi-Agent System. The framework adopts a five-agent architecture consisting of the Coordination Agent, Data Gathering Agent, Analysis Agents, Explainability Agent, and Audit Agent. The interaction protocols and workflows are formally defined using sequence diagrams and structured pseudocode. A large language model serves as the reasoning backbone for agent inference and natural language understanding, and human-readable explanation generation—enabling flexible parsing of unstructured financial data, semantic alignment of cross-organizational task requirements, and interpretable decision rationales. A self-optimization mechanism is constructed to adjust system parameters based on real-time performance feedback. Comparative experiments are conducted on a cross-border supply chain finance risk assessment dataset. Results indicate that the proposed system outperforms conventional methods in processing efficiency, decision accuracy, auditability, and explainability.

Keywords

Multi-Agent Systems; Explainable AI; Auditability; Large Language Models; Financial Analysis

References

[1] Li, Y., & Li, Z. (2023). A review of the application of artificial intelligence in the field of finance. Journal of Finance and Economics, 49(5), 1-16. [2] Ghaffari, G., & Lamo, Y. (2023). Explainable and Auditable AI in Finance: A Review. Journal of Risk and Financial Management, 16(3), 133. [3] Kansal, A. R., & Singh, Y. (2022). Algorithmic due process: a framework for auditable and accountable algorithmic systems. Journal of Information, Communication and Ethics in Society, 21(1), 18-36. [4] Pan, F., Chen, J., & Wang, Y. (2023). Large language models for finance: A survey. ACM Transactions on Intelligent Systems and Technology, 14(5), 1-21. [5] Wang, J., Zhang, C., & Wang, H. (2022). A survey on multi-agent systems for financial applications. Expert Systems with Applications, 207, 117978. [6] Battiston, S., & Caldarelli, G. (2023). Networks in finance: A review. Journal of Economic Literature, 61(3), 856-917. [7] Chen, X., Liu, W., & Thompson, R. (2023). Multi-agent systems in financial decision-making: A survey. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 53(4), 2345-2358. [8] Smith, J., Lopez, E., & Zhang, H. (2023). Interpretability in large language models for business process analysis. ACM Computing Surveys, 55(12), 1-42. [9] Wang, S., Rodriguez, M., & Kim, S. (2023). Explainable AI for automated compliance auditing in enterprise systems. ACM Transactions on Intelligent Systems and Technology, 14(6), 1-28. [10]Martinez, A., Brown, D., & Davis, M. (2023). Auditability in automated decision systems: Challenges and solutions. Information Sciences, 628, 156-172. [11]Noam, Y., & Sorin, S. (2021). Artificial intelligence and market manipulation. The Journal of Legal Aspects of Computing, 29(4), 543-571. [12]Kang, J., Zhao, M., & Wilson, P. (2022). Deep learning for process mining and anomaly detection. IEEE Transactions on Knowledge and Data Engineering, 34(8), 3567-3581. [13]Liu, W., Yang, T., & Patel, R. (2023). Blockchain-based immutable audit trail systems for regulatory compliance. Information Fusion, 92, 234-247. [14]Sun, H., Li, Y., & Chen, Y. (2022). A federated multi-agent system for privacy-preserving fraud detection. In 2022 IEEE International Conference on Big Data (Big Data) (pp. 3144-3153). IEEE.