Voino
Turns conversations into editable notes and a visual board.
- Category
- AI / Productivity
- Year
- 2026
- Role
- Product Engineer, Voice Model + Board Builder
- Status
- PROTOTYPE
- Links
- voino.vercel.app ↗
Voino is a voice-first productivity experiment that turns meetings, lectures and discussions into editable notes and a visual board.
The product is designed around a simple flow:
Speak or type → transcript → notes → visual board → edit, save and share.
It is built with Flutter for Android and web and follows a local-first approach.
Problem
Ideas captured during meetings and discussions often disappear into raw recordings, scattered notes or text that is difficult to organise afterwards.
Voino explores a different workflow: capture what was said, turn it into editable notes, then make those notes visually understandable.
Approach
The product separates capture, interpretation and visual organisation.
Users can speak into the application or provide text directly. The transcript can then be converted into notes, which are transformed into an editable visual board.
The system supports local extraction by default, with optional Gemini-based summarisation and hosted AI summarisation.
My Role
I worked primarily on the voice model / voice interaction layer and the board builder.
My work focused on the transition from spoken input into usable notes and from those notes into a visual workspace that users can edit rather than simply read.
Board Builder
The board turns extracted notes into an editable, Excalidraw-style canvas.
It supports:
- Automatically arranged note cards
- Topic grouping
- Relationships between cards
- Action-item cards
- Editing and repositioning
- Connecting cards
- Drawing boxes and circles
- Undo
- Save/open board files
- JSON-based board persistence
The goal was to make the generated structure a starting point rather than treating AI output as something the user should blindly accept.
Technical
Voino is built with Flutter for Android and web.
The project uses local extraction for notes, browser-based Whisper as an optional local transcription path, Gemini as an optional AI layer, and a Vercel serverless function for hosted summarisation.
The application is local-first, with transcripts, notes and boards remaining on the device unless the user explicitly enables an AI option.
Learning
The biggest lesson was that AI output should be treated as editable material rather than the final answer.
The board builder made this particularly obvious. The useful product isn't simply generating a mind map. It is giving the user a structure they can immediately correct, rearrange and make their own.