AI Internationalization + Localization Automation Infrastructure

Okcoin was expanding fast across more than 100 countries, and localized content had become the bottleneck. The product moved quicker than a manual localization process could keep up with.

The company was running on a bare-bones setup of an expensive translation management system (Memsource, then Phrase) that required manual work at every stage and wasn't connected to any of our content, development, QA, or testing tools. Localization was slow, and it was costing us customer retention, experience, and engagement.

I built an automation pipeline using AutoML, the Lokalise API, and Python scripts that handled 80% of company content and supported a product content pipeline for over 100 countries.

I led the development and fine-tuning of the company's first machine translation models to consistent BLEU scores above 80, which set the groundwork for releases without delays in 20 languages, and I defined the internationalization product roadmap against the company's expansion goals.

Replacing the old system meant picking a new one, then getting the whole company to actually use it. I ran it as a structured rollout:

• Vendor research: compared TMS options across cost, ease of use, and scalability
• Executive buy-in: pitched OKX's CEO and management team on what automating each stage would cost, and what it would buy us
• Org-wide rollout: once leadership signed off, presented the same case to management teams, then to 700+ people whose day-to-day work it would touch directly, with a customized automated workflow designed per team
• Phased execution: the main content pipelines, the website and mobile app codebases, converted in about 6 months; the rest of the org's workflows followed over the next 6 months

I served as interim lead, training 15+ newly hired members of the globalization team on project management, Agile best practices, and Jira standards.

A year into the rollout, we'd automated 80% of every step across every content pipeline except legal, serving 100+ markets, and saved thousands of hours a month along with a meaningful chunk of what we'd been paying our previous vendor. The new pipeline also raised user engagement 37% and increased translated-content satisfaction 25%, with on-time releases in 20 languages.

Automated checks also caught roughly 400 i18n issues before they ever reached production each month. Across a 20-language, heavily regulated product, that saved an estimated 600 to 1,400 engineering, QA, and compliance hours a month, the equivalent of 4 to 9 full-time engineers freed up for core product work instead of localization fixes.