Python-based intelligence layer for SWIFT messaging systems using LLMs to predict routing and compliance failures
Sappa, Ankita
Sappa, Ankita
Citations
Altmetric:
Authors
Other Names
Location
Time Period
Advisors
Original Date
Digitization Date
Issue Date
2025-10-08
Type
Conference paper
Genre
Keywords
Compliance monitoring,Financial message intelligence,GPT-based analysis,Large Language Models (LLMs),Python-based financial systems,Real-time anomaly detection,Regulatory risk detection,Routing failure prediction,SWIFT messaging,Transaction routing optimization
Subjects (LCSH)
Citation
A. Sappa, "Python-Based Intelligence Layer for SWIFT Messaging Systems Using LLMs to Predict Routing and Compliance Failures," 2025 International Conference on Next Generation Computing Systems (ICNGCS), Coimbatore, India, 2025, pp. 1-8, doi: 10.1109/ICNGCS64900.2025.11183123.
Abstract
In recent years, SWIFT message routing efficiency and adherence within the financial domains are of utmost importance for the global transaction systems' integrity. This paper describes an artificial intelligence layer implemented in Python for detecting and anticipating compliance and routing renegade procedures within SWIFT messaging frameworks, powered by Large Language Models (LLMs). We developed a hybrid pipeline that integrates natural language embeddings with semi-structured text fields, utilizing over 4.2 million historical SWIFT messages. The designed system enables the real-time tracking and analysis of regulatory violations in transactions, enhancing the system's ability to minimize false positives and lag while maintaining high levels of interpretability. We apply various architectures of LLMs, such as BERT, GPT-Neo, and a fine-tuned domain-specific variant of GPT, to compare with traditional sequence models. The analysis reveals improvements in predictive performance, with the best model achieving 92.4% accuracy and an AUC of 0.96 in detecting compliance failures. Forecasting routing failure showed a 31% better error rate than the rule-based benchmarks. Furthermore, the insights provided by the LLMs' attention mechanisms concerning compliance decision processes enhance the transparency required for auditability behind some structured message fields, exposing critical messages. This research demonstrates the potential for integrating language models into the financial message routing backbone, particularly in proactive regulatory anomaly detection and compliance risk management. The system provides a foundation for further development, such as learning from evolving compliance benchmarks and operating within federated environments. © 2025 IEEE.
Table of Contents
Description
Click on the DOI link to access this article at the publishers website (may not be free).
Publisher
Institute of Electrical and Electronics Engineers Inc.
Journal
Book Title
Series
2025 International Conference on Next Generation Computing Systems, ICNGCS 2025
2025-08-21 through 2025-08-22
Coimbatore
213525
2025-08-21 through 2025-08-22
Coimbatore
213525