#SuccessStories - Zero Regressions: How Sia Broke NARSA's Legacy Monolith into Seven Microservices with Claude Code
Back to Insights

September 28, 2026

#SuccessStories - Zero Regressions: How Sia Broke NARSA's Legacy Monolith into Seven Microservices with Claude Code

1. Context


Legacy monoliths with heavy technical debt and no automated tests are risky to modernize: any change can break behavior nobody has documented. AI coding agents now make it possible to analyze such code, map dependencies and rebuild it faster, provided the work is driven by specifications and proven by tests.



NARSA's backend application was a legacy Spring Boot monolith of around 392 Java files, with heavy technical debt and no automated tests.



2. Client Challenge


NARSA needed to:



  • Transform its legacy monolith into maintainable microservices.
  • Avoid any regression in behavior.
  • Build the automated tests the application had never had.


3. What Sia Did ?


Using Claude Code as an engineering agent, Sia accelerated the transformation into seven microservices through a controlled software factory:



  • Claude Code-assisted reverse engineering: analyzing the monolith, mapping dependencies, identifying technical debt and defining clear bounded contexts, with specifications written before any code.
  • Skills-based workflow: dedicated skills for architecture mapping, service design, tests and microservice extraction, creating a repeatable cycle for each service: specification, plan, tests, code, review, validation.
  • Definition of Done: each microservice had to prove parity with the original monolith: green characterization tests, more than 80% coverage, health check up, no missing behavior, no hardcoded secrets, an independent review and final validation by NARSA.


4. Results & Outlook


The transformation delivered:



  • 7 of 7 production-ready microservices.
  • Zero regressions, with behavioral parity proven by tests.
  • 513 green tests, from zero initially, and 89% to 96% test coverage across services.
  • The largest 'god class' decomposed from 2,188 to 145 lines.


By combining AI coding agents with specification-driven engineering and rigorous validation, Sia helps organizations modernize legacy code quickly and safely.