Watch it watch you.
Mimic the sign.
Know instantly.
A looped clip of a real signer on one side, your camera on the other. Pass or try again — with a confidence score, never a guess.
Sign it
to search it.
Forgot a sign's name? Show it to your camera and get the closest matches back — or search by text and see a real signer demonstrate it.
Five minutes a day
is what sticks.
Spaced repetition resurfaces each sign right before you'd forget it. A quiet streak keeps you honest — no guilt mechanics.
teach. bridge.
stay free.
70 million people sign as their first language. The tools to learn it, or bridge it, are scarce, costly, or closed. This one isn't.
Four ways in.
One product, four jobs — from your first letter of the alphabet to looking a sign up by showing it to your camera.
21 points.
30 frames a second.
On your device.
Folang reads the landmark coordinates of your hand — not the pixels of your face. The model is small, fast, and private. Your webcam feed never leaves the browser.
Builtsothepeoplewhoneeditmostneverpay,arenevertracked,andareneverpromisedsomethingthatdoesn'twork.
Five promises.
Non-negotiable.
Five commitments we won't trade away — they shape every technical decision, not just the copy.
Privacy-first
All recognition runs in your browser. No video upload, no cloud inference, no surveillance — ever.
Accessibility-first
Captions, keyboard nav, screen-reader support. Nothing relies on audio alone. It's a requirement, not polish.
Deaf-community-led
Deaf signers shape the data, validation, and direction from day one — credited, and listened to.
Honest about accuracy
Every model ships a model card with its real accuracy and known failure modes. The UI never implies more certainty than it has.
Genuinely open
Permissive license, public roadmap, real contribution paths. No bait-and-switch to a paid tier, not now, not later.
Honest about what it can't do.
Overpromising is how tools like this lose the community's trust. So we're precise about scope — in the docs, and in the product itself.
Folang is
Folang is not
Each phase ships something real and de-risks the next. We're currently shipping Phase 2–3: a working alphabet Learn mode.
Foundation
Lock the language and scope. Repo, license, code of conduct, ethics note — and reach out to Deaf collaborators.
Camera + landmarks
Live 21-point hand tracking at 30fps in the browser — working on a mid-range phone, not just a flagship laptop.
Fingerspelling classifier
A–Z static recognition: camera → landmarks → letter + confidence. The tractable “hello world” of sign ML.
Learn mode v1
Alphabet lessons, the mimic-and-feedback loop, a progress streak. The first genuinely usable build.
Dynamic isolated signs
A sequence model over 50–100 common signs, validated on unseen signers. Where the real ML work lives.
Accounts, content, accessibility
Optional auth and progress sync, a content system, and a full WCAG 2.1 AA audit before release.
Translate mode, experimental
Fingerspelling → text first, then a small fixed vocabulary. Clearly labelled limits, the whole way through.