# TeacherDox

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Sandeep Kumar](https://cerebralvalley.ai/u/teacherdox)
- **GitHub:** https://github.com/sandeepkumareeri/teacherdox-core
- **Demo video:** https://www.youtube.com/watch?v=BRQtIwP1mf4
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3798

TeacherDox reads the board and writes the teaching diary.

Across the Global South, most teachers work without computers, ERPs or school software: just a board, a register and a pen. So every lesson is written twice: on the board, then wiped off and copied again by hand into a teaching diary. That costs a teacher 30–45 minutes every day, and the result is records that are slow to write, hard to check and easy to lose.

TeacherDox is WhatsApp-first teaching records infrastructure for everyone who teaches: schools, colleges, universities, tutors and trainers. It is live in production on WhatsApp in India, with school pilots starting in Kerala and Andhra Pradesh.

How it works: after class, the teacher snaps the board on WhatsApp and, if they like, adds a voice note or types a line. No app, no forms. Meta's models find the board in the photo, hear the voice note, read the board and write a structured diary entry: class, subject, topic, what was taught and homework, each field with a confidence. A second model call checks every field against the photo. If a field is unclear (on our real Andhra Pradesh board, the class number), TeacherDox asks the teacher instead of guessing. It never invents. In the evening, one tap signs the day. Headmasters and education offices see the diary live.

What makes it infrastructure and not a chatbot: every signed entry becomes an open .tdx record, signed with Ed25519, linked in a daily hash chain and sealed with a Merkle root. Anyone (a headmaster, an inspector, a ministry) can verify it in a browser with no special software, and a signed record can never be silently edited; corrections are new, linked records.

Built during the hackathon (Oct 27 – Nov 4), all in the public repository with dated commits:
- the .tdx 2.0 open teaching-record spec, JSON Schema and conformance test vectors
- a zero-dependency trust layer (canonical JSON, signatures, hash chains, Merkle seals, countersignatures)
- the Record Engine on Meta's models (locate → listen → read → check), with confidence routing, visible fallback and per-step provenance
- packs for 7 languages, 6 institution types, 5 countries and 4 record policies
- a REST API with a durable queue, a WhatsApp Cloud API channel, a CLI and a public in-browser verifier
- an evaluation harness with published results: Meta Model API 81% field accuracy vs Gemini 77% and Muse Spark Open 69%, 98% on boards
- 98 tests, CI, type checks, a threat model and release v0.1.0

The TeacherDox WhatsApp product existed before the hackathon; the open infrastructure above is what we built for it.

## More from Meta Global AI Developer Hackathon

- [Adrien Lachambre](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3795)
- [Fieldglass: The Reality Compiler](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3796)
- [ClimateAdapt Lab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3797)
- [NegotiateAI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3799)
- [Ansh](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3800)
- [PriceCharm](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3801)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3798. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt