Transept AI
AI Translation workspace with Translation Memory
Visit Website
July 18, 2026
Localizing an AI product into ten languages
Here goes an absolutely wild story on how we translated ~6,300 strings in the application and landings into 9 languages - Localizing an AI product into ten languages — Transept.
Translated well, with context and puns preserved. But had to improve internationalization engine a lot to serve all the unique features of languages. And all ways of ISO quirks, breaking and full-width spaces, and stylistic differences. As well as LLMs literally changing their character in different languages, too.
1 Comment
The interesting challenge isn't translating an AI product into multiple languages—it's preserving the same user trust and product identity across cultures while the underlying models behave differently. I'd keep validating whether localization improves adoption because users understand the product better or because they feel it was genuinely built for their market.