Kyutai — Invincible Voice
Taking a research prototype to a product ready to be demonstrated, released and picked up as open source, in under two months.
- Sector
- Artificial intelligence · Accessibility
- Engagement
- Prototype → production
- Timeline
- November 2025 → January 2026
- Expertise
- Product design & complex interfaces · Product architecture & engineering
Context
Kyutai is a French artificial intelligence research lab. In late 2025, their teams were working on Invincible Voice, an interface designed to let people who have lost the use of speech — in particular through ALS — communicate more easily with those around them.
By the time Kyutai got in touch, the core of the product already worked. But it was still a prototype: the interface had been built to test usage, and the codebase had to evolve before the project could be released more widely.
Two deadlines overlapped. In the immediate term, several demonstrations were scheduled, and the interface had to feel like a finished product straight away. Beyond that, Invincible Voice was meant to be published as open source: performance, code quality, authentication and the ability for others to pick the project up then mattered just as much as the visible experience.
The purpose of the project mattered to us: putting our product and engineering skills behind a tool meant to make communication more accessible.
Our approach
The first priority was set by the calendar: make the experience presentable without waiting for the full technical rework of the product.
Sylvain therefore proposed a visual redesign within a few days. Valentin implemented it immediately, so that Kyutai could run the scheduled demonstrations on a finished interface.
Once that deadline was secured, we took the codebase further. The goal was no longer only that the product should work, but that it should be publishable, maintainable and reusable by other developers. So we set up an independent repository and reworked the frontend, the backend and the infrastructure around that new goal.
Performance was part of the picture too. The audio streams in particular were reworked to run as streaming rather than prefetch, making the experience noticeably more responsive.
From a research prototype to a product others can pick up
What we delivered
Within a few weeks, we delivered a new version of Invincible Voice.




A fully reworked user experience
A complete visual redesign, implemented in time for the first demonstrations.
A new frontend and backend codebase
An independent repository and a rewritten application core, ready for release and for other developers to take over.
A more responsive audio experience
Audio streams moved to streaming, improving the responsiveness of the application.
The pieces a usable product needs
Internationalisation, email and Google authentication, user management and the supporting infrastructure.
The final version of our engagement was delivered in January 2026, less than two months after the first conversation.

What Kyutai said
Faced with very tight time constraints, Vezero rose to the challenge brilliantly, combining responsiveness and professionalism. The quality of the UX and the fidelity of the implementation genuinely impressed us, all the more so given the deadlines!
Do you need to take a prototype to a real production version?