Do, don’t just watch
Every lesson has you redo each sign straight away, then produce it from memory. Each sign has two review cards: recognise (video to word) and produce (word to sign).

Pl. 02 / 16Case study
An app to learn French Sign Language ten minutes a day, by redoing every sign in front of a camera mirror.
Presentation film made for this portfolio. The trajectory is the one the app measured on the reference video of the sign “Merci” (Laura Jauvert, CC BY 3.0).
For English or Spanish there are dozens of apps. For French Sign Language there are excellent video dictionaries and a few courses, but there you watch signs, you almost never make them. I wanted an app for people who want to learn LSF ten minutes a day, for instance to talk with a deaf relative.
A sign language is learned with your hands: the app had to make people produce the gesture, not just recognise it, and correct it without the learner’s video ever leaving their phone. It also needed a warm identity that looked neither like clipart nor like a school app.
I reviewed sign language apps and the available figures: about 100,000 deaf people have LSF as their main language and 250,000 people use it (Université de Rouen). A 2024 study (Morett et al.) shows that a sign you have reproduced is remembered better than one you have only watched.
Every lesson follows the same short loop: watch the sign, redo it, recognise it among four options, sign it from memory, then review it at the right time with spaced repetition (FSRS).
MediaPipe finds the hands and shoulders in the browser. The points are normalised to the upper body, then the trajectory is aligned with the signer’s using DTW. The references were precomputed from the 99 videos in the course.
On 20 October I presented three complete directions, each with a logo, palette, typefaces, an app screen and a post: Lumière du jour, Le geste and Atelier papier. I chose Atelier papier, keeping Bricolage Grotesque, the beta’s typeface, for the word “paume”.
The app can be installed and works offline. It is hosted on Cloudflare Pages; a few functions and a D1 database collect feedback and, only with consent, anonymous statistics.
















Every lesson has you redo each sign straight away, then produce it from memory. Each sign has two review cards: recognise (video to word) and produce (word to sign).
The gesture is analysed in the browser and no image is sent anywhere. Progress stays on the phone, with no account and no cookies.
Every video shows a real person, credited with their name and licence (Wikimedia Commons). I didn’t want avatars or drawn signs: a wrong sign would be learned just as well as a right one.
LSF is a language of the eyes, learned with the hands. The symbol is a cut-out hand printed in two riso inks. I shifted the pink almost vertically so it doesn’t streak the fingers at small sizes, and the eye is cut through both sheets.
Signer videos and photos always sit on a plain soft blue, never on the beige paper or a texture, so the hands stand out.
This project taught me to think about identity and product together: the colours, type and shadows in the guidelines are the same variables as in the code. The next step is written into the guidelines: have deaf people review the direction and the course before a wider release, since the eye in the palm is a tribute to their language.
Colours sampled from the project visuals.