Building LinguaLens: A Floating Translator for Real Desktop Work

Here is a scene repeated in offices across Pakistan every day: an engineer receives a German equipment manual as a PDF, a researcher opens a French paper, a supplier's specification arrives in Dutch. The text is understood the slow way — copy a paragraph, switch to a translation website, paste, read, switch back, find your place again, repeat forty times. The knowledge is on the screen; the friction is in reaching it. We built LinguaLens AI to remove that friction.

LinguaLens is a Windows desktop application whose centrepiece is a floating lens: a small, transparent, always-on-top window you drag over whatever you are reading — a browser page, a PDF, a Word document, an email. Draw the lens over a paragraph and it reads the text with OCR (optical character recognition, using the open-source Tesseract engine), then translates it into Simple English or Urdu. The lens can be moved, rotated, resized and made more or less transparent, so it sits on your document the way a magnifying glass sits on a page.

Three workflows cover most real use. Region capture is the magnifier mode just described. Live mode watches a region continuously and re-translates when the text changes — useful for scrolling documents, dashboards and chat. Clipboard mode translates whatever you copy with a single shortcut. Everything is driven by global keyboard shortcuts, because a tool that demands mouse trips to its own window has missed its own point.

Two design decisions matter more than the features. The first is that translation runs on your own AI account: you paste your own Google Gemini or OpenAI key into the settings — Gemini offers a free tier — and the app talks directly to that provider. There is no AstolixGen server in the middle reading your documents. The second is where things are stored: your key is encrypted using Windows' own data protection (DPAPI) and never leaves your machine except to call your chosen AI provider, and your translation history lives in a local database on your PC that you can search, export or wipe at any time.

We also learned the honest limits of OCR while building it. Recognition is excellent on clean, reasonably sized text and noticeably weaker on tiny fonts, low-contrast scans and decorative typefaces — which is why the app shows you the recognised text and lets you correct it before translating, instead of silently translating a misreading. Urdu output renders right-to-left properly, and for spoken output the app uses Windows' installed voices and tells you plainly when no Urdu voice exists on your system rather than pretending.

LinguaLens is open source under the MIT license — the code is public on our GitHub (github.com/asifaliattari/lingualens-ai) — because a tool that asks for your trust should show its workings. It is free to download and use with your own AI key. If you try it, we genuinely want the bug reports and the hard cases: the scanned Urdu newspaper it misreads, the German table it mangles. Tools get good by being used in the open, and translation that respects its reader's language — including Urdu, too often treated as an afterthought — is worth doing properly.

If you want to try it, the fastest path takes about ten minutes. Install LinguaLens on a Windows 10 or 11 PC, then add your own translation key in Settings → AI translation provider — a free Gemini key from Google AI Studio is enough to start, since Google offers a free tier for developers and small projects. Then try clipboard mode first: copy any paragraph from a document and press Ctrl+Shift+T to translate it on demand. Once that works, try region capture on a scanned page, and remember the app shows you the recognised text so you can correct it before translating. Your history stays in a local database on your PC, where you can search it, export it as TXT or CSV, or wipe it.