Case Study: AI-Assisted Legacy Modernization of a Loan Servicing Platform

Banking

AI-Assisted Legacy Modernization of a Loan Servicing Platform

The Challenge: A Core System Nobody Could Safely Change

A mid-sized bank ran its loan servicing operation on a desktop application first built more than twenty years ago. It had grown to roughly 600,000 lines of code, with business rules spread across screens and database stored procedures, almost none of it documented. Only a handful of engineers still understood it and two were nearing retirement, so every regulatory change was slow and risky to make. A conventional rewrite had been estimated at three years, which the bank could not justify, and the platform stayed in place while the risk of keeping it grew.

Our Solution: Coding Agents Under Engineering Control

Lydatum ran the modernization as an engineering program in which AI coding agents did much of the code reading and translation, directed and reviewed by our engineers and the bank's own team:

• Automated Inventory & Documentation: Agents analyzed the whole codebase and produced a structured inventory of every screen, workflow, business rule and data operation, which the bank's subject matter experts reviewed and corrected.
• Target Architecture & Migration Plan: We designed a modern web architecture and sequenced the migration module by module, so each phase delivered something the business could verify.
• Agent-Assisted Translation: Coding agents translated each module into the new stack to agreed patterns. An engineer reviewed every change, and each one had to pass the bank's security scanning and test coverage standards.
• Automated Parity Testing: For each migrated function, tests ran the same inputs through the old and new systems and compared the results, which gave objective evidence that behavior was preserved.

Data moved through repeatable, reconciled pipelines, and the two systems ran in parallel for a full servicing cycle before cutover. The bank's engineers worked alongside ours throughout and now own both the new platform and the AI-assisted way of working that produced it.

The Impact: Three Years of Work Delivered in Fourteen Months

The new platform went live on schedule and the legacy system was retired shortly afterwards.

The program took 14 months against the 36 originally projected. The bank now has a documented, tested record of how its servicing rules work, which it did not have before. Regulatory changes that used to take a quarter are delivered in weeks, and the platform no longer depends on a few individuals.

Services Delivered: AI-Assisted Legacy Modernization, Agentic SDLC & Engineering Workflows, AI Code Governance & Guardrails, Data Pipelines & Integration

Technologies Used: Claude Code, C#, .NET, React, TypeScript, SQL Server, PostgreSQL, GitHub Actions, Playwright