Regrello Corp. – an AI-focused startup in supply chain management, acquired by Salesforce in late 2025 – needed a complete internationalization effort leveraging large language models. I led this initiative and evaluated the quality of translations from English into six target languages: French, German, Chinese, Spanish, Korean, and Thai. The developed plugin system enables automatic translation of all user inputs, significantly simplifying the product's international scaling.
A plugin system for the Go backend's GraphQL resolvers and mutations formed the core of the solution. New user inputs are automatically translated via LLM (Google Gemini and gemma3) and cached in Redis. Translations are orchestrated through Temporal workflows, for which I implemented corresponding queues and workers in Go. Since plaintext user content could not be stored in the Temporal queue for security reasons, I developed a performant encryption system for Temporal. The backend was structurally improved and decoupled, making it more maintainable and comprehensible.
