Kensi
Multilingual voice-to-text for macOS and iOS that turns speech into corrected text wherever the cursor is, without outsourcing transcription.
The problem worth solving
Most dictation products force a compromise between convenience, accuracy and privacy. They interrupt the writing flow, perform speech recognition remotely or make the user move text from a separate transcription window into the application where it belongs.
What I built
Kensi reduces dictation to a direct interaction: hold a key on macOS or use the iOS keyboard microphone, speak naturally and receive text at the active cursor. Speech recognition remains on-device and supports multilingual use rather than assuming an English-only workflow.
The product combines local transcription with custom vocabulary, searchable history and optional language-model correction using credentials controlled by the user. Its shared Swift foundation keeps the interaction consistent while respecting the different capabilities of macOS and iOS.
What the product does
- System-wide macOS dictation through a global hold-to-talk keyboard shortcut.
- An iOS keyboard extension that brings the same workflow into mobile applications.
- Multilingual, on-device speech recognition with no remote transcription service.
- Custom vocabulary, searchable history and optional user-controlled text correction.
Why the design matters
The important choice in Kensi is where the work happens. Speech recognition stays on the device, while native cursor insertion puts the result directly into the user’s current application instead of creating another destination for their words.
My role
Ciprian created Kensi and leads its product, interaction design and native engineering across the macOS application, iOS application and keyboard extension.