LLM-Powered Banking Modernization: A Human-Centric Approach to Legacy Code Transformation

Abstract

Many financial institutions are at a crossroads—still running on decades-old COBOL systems that eat up IT budgets and slow innovation. Traditional modernization approaches have been risky, expensive, and often unsuccessful. It’s time for a smarter, more scalable solution.

This talk introduces a cutting-edge framework that blends the power of Large Language Models (LLMs) with human expertise to modernize legacy banking code—securely, efficiently, and with a deep respect for regulatory and operational demands.

I’ll walk through a real-world-tested architecture that addresses the toughest challenges in banking modernization: behavioral equivalence, complex domain-specific code, and seamless integration with legacy infrastructure. The framework leverages LLMs with extended context windows, domain-specific fine-tuning, and a role-based human-in-the-loop system for oversight and compliance.

Our case study—modernizing a major COBOL-based transaction system into containerized microservices—shows just how transformative this approach can be. The results? Dramatically reduced technical debt, faster feature delivery, and real, measurable ROI.

This session is designed for tech leaders in financial services looking for an actionable blueprint to future-proof their core systems using AI—without compromising stability or compliance.