Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
Education Pilot: Sector 1 Bucharest (Draft)
Status: exploratory. A contact has offered a local connection to the office of the deputy mayor of Sector 1, Bucharest (education and schools portfolio). This doc holds the outbound reply and the one-page pilot brief to forward once there is interest.
The honest position (read before pitching)
- Our signing avatars speak ASL (American Sign Language), not LSR (Limba Semnelor Române / Romanian Sign Language). These are different languages. Never pitch the current feature as an accessibility tool for Romanian deaf students.
- What we can honestly pitch today: 3D learning assistants for regular classes, plus a unique English-learning loop (avatar signs and fingerspells English words; students practice back at a webcam that grades their handshape on-device).
- The signing engine is language-agnostic: signs are authored as data (src/sign-dictionary.js), so LSR is content work, not an engine rewrite. Building LSR credibly requires native LSR signers and educators reviewing every sign; a school partnership is exactly how to get that review.
- Privacy story is strong and real: the practice loop (/sign-mirror) runs fully on-device with no network call, and the recognition lane uploads hand landmarks only, never video. No accounts are required for any of the learning surfaces.
Reply draft (send to the contact)
Hey, thanks for bringing this to me. Yes, we're open to it, and I'd like to explore it seriously.
One thing I want to be upfront about so we set the right expectations with the school: our signing avatars currently speak ASL (American Sign Language), not LSR (Romanian Sign Language). Those are different languages, so I don't want to walk in promising an accessibility tool for deaf students that isn't there yet.
That said, here's what we can do today, and what I'd propose:
Phase 1 (ready now): a regular-class pilot. 3D agents as learning assistants, plus an English-learning angle that's genuinely unique: kids type an English word, a 3D avatar signs and fingerspells it, and they can practice signing back at the webcam, which grades their handshape in real time. Privacy-wise it's built for exactly this setting: the camera video never leaves the device, only hand landmark points are processed, and the practice loop runs fully on-device. No accounts needed.
Phase 2 (co-development): true LSR support for the special-needs group. Our signing engine is language-agnostic, so adding Romanian Sign Language is content work, not a rebuild. But we won't do it without native LSR signers and educators reviewing every sign. If the school can connect us with LSR-fluent staff or interpreters, we'd build the LSR vocabulary with them. That makes the pilot a real partnership rather than us dropping software on them.
If your contact is open to it, a 30-minute call with whoever runs the education programs would be the right next step. I can put together a one-page brief in advance. And happy to have you make the local connection, that would help a lot.
One-page pilot brief (forward after interest)
three.ws education pilot: 3D learning assistants and signing avatars
What three.ws is. A web platform for 3D AI agents: animated avatars that talk, teach, and sign, running in any browser with nothing to install. Free to use for the surfaces in this pilot; no student accounts required.
What we propose.
Phase 1: one or two regular classes (4-6 weeks).
- 3D learning assistants. A teacher-configured 3D avatar embedded in class materials that students can ask questions in natural language.
- English + sign language practice. Students type an English word and a 3D avatar fingerspells and signs it (three.ws/sign-language, three.ws/asl-alphabet). In the mirror room (three.ws/sign-mirror), the webcam watches the student's hand and grades the handshape live, naming the finger that is wrong. Works on ordinary school laptops.
- What we measure. Engagement (voluntary practice time, alphabet completion), teacher-reported usefulness, and a before/after fingerspelling reading test.
Phase 2: co-developing Romanian Sign Language (LSR) support with the special-needs group.
- Our signing engine is not tied to any one sign language; vocabularies are authored as data and reviewed by fluent signers.
- With LSR-fluent staff or interpreters from the school, we would author and review an initial LSR vocabulary, giving Romania its first browser-native LSR signing avatar, built with the community that uses it.
- The special-needs group and its educators define the vocabulary priorities; we ship review tools and iterate weekly.
Privacy and safeguarding.
- The handshape practice loop runs entirely in the browser: no video frame, landmark, or score is ever uploaded, and there is no network call in the practice loop.
- The optional sign-reading feature uploads anonymous hand landmark coordinates only, never video or images.
- No student accounts, no personal data collection, no ads, no tracking of minors. GDPR posture available on request.
What we need from the school.
- 1-2 participating teachers and a 30-minute onboarding call.
- Ordinary laptops or tablets with a webcam and a browser.
- For Phase 2: access to LSR-fluent staff or interpreters for sign review sessions.
Cost. The pilot is free. We fund the infrastructure; the school contributes feedback and, in Phase 2, sign-language expertise.
Next steps checklist
- Send the reply above to the contact.
- If interest: schedule the 30-minute call, forward the brief.
- Phase 2 groundwork: scope an LSR vocabulary format and a signer review flow on top of src/sign-dictionary.js (do not start authoring signs without native signer review lined up).