# Lotus

- **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:** [Minh Thang](https://cerebralvalley.ai/u/MinhThanghus)
- **GitHub:** https://github.com/MinhThanghus/Conceptlab
- **Demo video:** https://youtu.be/GydaNWnYqE8
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/653

ConceptLab AI turns a messy whiteboard, a textbook page or a practice video into an interactive, source-cited lesson in minutes.



THE PROBLEM

Some of a lesson's best explanations end up on a whiteboard that is erased at the bell, or sit in a static textbook figure. Teachers rarely have time to turn either into something students can explore, and many students learn on cheap phones over slow networks.



WHAT CONCEPTLAB DOES

A teacher uploads what they already have: a phone photo of the board, textbook pages, PDFs, practice videos or measured values. Meta's Muse Spark reads the material and extracts the topic, formulas, variables, steps and anything missing. It also tells the teacher what it left out, such as a crossed-out formula. The teacher reviews everything in plain fields. ConceptLab then builds the lesson:

1. 3D and 2D simulations bound to tested physics models (pendulum, projectile, spring, compound interest), with sliders tied to the real formula.

2. Live graphs, clickable diagrams whose parts are found by SAM 3.1, and step-by-step slides that highlight the exact source line.

3. A clean, proofread copy of the handwriting: Muse Spark reads each line, Muse Image redraws the sheet, and Muse Spark checks every word.

4. Self-check quizzes marked on the server, so answer keys never reach the browser.

5. An AI tutor that answers only from the lesson's sources, with citations. Students can type or ask out loud (Muse Voice Transcribe).



The teacher shares the lesson with a 6-character class code, and students need no account. Everything works in Vietnamese and English, and Lite mode keeps lessons usable on low-end phones, on slow networks and offline.



Two spaces share one platform: School2Lab for grades 6-12 (including physical education, with SAM-tracked practice videos) and Paper2Explorer for university students reading research papers.



TRUST BY DESIGN

- Every AI-made part is labelled until the teacher checks it.

- Questions reach students only after the teacher approves them.

- 23 automatic checks run before a lesson can be shared.

- Numeric answers are recomputed by the app's own formula engine.

- Labs with media showing students switch to Meta's Standard (private) mode automatically.



BUILT WITH

React 19 and Three.js on the frontend; FastAPI and SQLite on the backend; the Meta Model API (Muse Spark, Muse Wonder, SAM 3.1, Muse Image, Muse Voice Transcribe). Open source, with 1,001 frontend and 565 backend automated tests.

## More from Meta Global AI Developer Hackathon

- [Samir Baral](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/650)
- [ScoutIQ](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/651)
- [Innovators](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/652)
- [Sheepin Team](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/654)
- [KARTS AI ENTERPRISE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/655)
- [Team FixLens](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/656)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/653. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt