Automated I18n Quality for Enterprise Platforms
Over the past decade, our fintech platform has grown through numerous feature expansions and acquisitions, resulting in a patchwork of development methodologies, tech stacks, and ad hoc internationalization code tailored to specific markets.
Inspired by recent research into AI-powered “i18n self-healing”, we set out to train machine learning models on both best practices and known anti-patterns in internationalization code. Our goal: to detect problematic implementations and recommend CLDR/ICU-compliant improvements, especially for issues not normally handled by standard i18n libraries such as inconsistent postal address, phone numbers, and other personal information.
In this talk, we’ll share our journey to date implementing custom AI-driven i18n anti-pattern scanners in our localization infrastructure. We’ll demonstrate examples of detected issues and highlight the cases where this approach excelled compared to alternatives. Attendees will leave with practical insights into defining requirements, training specialized models, and embedding AI-enhanced i18n quality checks into workflows for complex, legacy codebases.