Symbolic and Explainable Regulatory AI for Finance Innovation (SERAFIN)
Financial regulations are increasingly complex, often overlapping and conflicting, making compliance difficult for institutions and regulators. This creates uncertainty for professionals, investors, and the general public. How can financial actors navigate these norms, and how can AI provide trustworthy explanations?
SERAFIN will develop a hybrid AI system that combines the reliability of symbolic methods with the adaptability of large language models (LLMs). The system will process financial regulatory texts—focusing on sustainability-related norms—and use argumentation frameworks, dialogue protocols, and retrieval-augmented generation (RAG) to resolve conflicts and provide personalized, transparent explanations. The project is structured around three pillars:
- Financial regulation—collecting and analyzing regulatory texts,
- Symbolic AI—formal argumentation and conflict resolution, and
- Sub-symbolic AI—fine-tuned LLMs orchestrated as multi-agent systems.
Together, they will enable robust compliance tools that support regulators, financial institutions, and citizens. As a proof of concept, SERAFIN will implement and test its system on EU sustainable finance regulations (MiFID II, SFDR, EU Taxonomy). The outcome will be both academic contributions to hybrid AI and practical tools for efficient, explainable compliance management
SERAFIN is a CORE project of the University of Luxembourg, funded by FNR (2025-2028).