Lingolense — AI Translation & Localization
Translate designs into 40+ languages in one click.
Hey IH,
Long-time lurker, first real post. We're launching Lingolense on Product Hunt next Tuesday, and I wanted to share the story here first — partly to give back to a community I've learned a ton from, partly because the hardest problem we solved might be interesting to anyone building design or i18n tooling.
The itch:
Localization is the step everyone treats as an afterthought, and it quietly costs designers days on every project. A design gets finished, then "now do it in Arabic and German" — and it becomes a week of copy-pasting Google Translate, rebuilding RTL layouts by hand, and fixing buttons that broke because German ran ~30% longer than the English we designed for. Every tool we tried "supported" RTL by swapping the font and right-aligning text. That's not localization — that's a broken layout in a different language.
What we built:
Lingolense does translation and localization inside the editor (Figma, FigJam, Slides, Pixso, Penpot). Select text, pick a language, translate — 40+ languages, routed between DeepL and an LLM depending on context. The part we're proudest of is true RTL mirroring: it duplicates the frame, reverses auto-layouts, swaps padding, flips directional icons (but leaves non-directional ones alone), and right-aligns text — original untouched. It also lags text overflow per language before it ships, and exports CSV/JSON/XLIFF for devs.
What I've learned so far:
- The 80% (translate the text) is easy; the last 20% (make the layout correct in the new language) is the entire product.
- "Which icons are directional?" is a deceptively deep problem.
- Talking to designers who localize daily changed the roadmap more than any amount of building in a vacuum did.
The ask:
We launching next Tuesday. Free plan, no card. If you localize designs — or you've built anything in the i18n space — I'd genuinely love your feedback on the approach before then, and your support on launch day would mean a lot. I'll drop the link in the comments when we're live. Happy to answer anything about the RTL logic, the DeepL-vs-LLM routing, or the go-to-market. AMA.
2 Comments
The strongest part is recognizing that translation itself isn't the hard problem—making the design survive the translated language is. The RTL mirroring and overflow checks sound especially useful because they address the actual layout failures rather than just producing translated copy.
Thanks — you've put your finger on exactly the reframe that took us longest to internalize.
The overflow check is the one that quietly saves the most pain, because the failure is invisible until it isn't. English "Save" is 4 characters; German "Speichern" is 9; Finnish can be worse. You don't find out until a button clips or a nav wraps in production — usually reported by someone who doesn't read the language. So we run it before export and flag the specific nodes at risk per language, rather than making the designer eyeball 40 locales.
RTL was the harder engineering problem but the more satisfying one, precisely because "swap the font and right-align" is the industry default and it's just wrong. Mirroring the layout — reversing auto-layout direction, swapping padding, flipping only the directional icons — is what makes an Arabic screen actually feel native instead of like English wearing a costume.
Curious, since you mentioned building in the space: on your end, was overflow or mirroring the bigger source of layout bugs? They fail so differently that I'm always interested in which one bit people first.