# -Y-

- **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:** [Abhijith Reddy D](https://cerebralvalley.ai/u/whatupdanger)
- **GitHub:** https://github.com/Abhijithreddydasari/ask-Y
- **Demo video:** https://youtu.be/yeKe_ICI8Vg
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4674

AI is advancing faster than ever, yet education is still one-size-fits-all: nothing tracks how you personally understand a subject. And a chat box isn't enough for complex ideas: arguments that need diagrams and working get flattened into text, and you can't easily show the AI what you're thinking or share the context it needs.

Y is a whiteboard-native AI tutor that learns how you think. The biggest ideas have been worked out on boards, so that is where Y lives. You write with a pen or speak, and Y teaches back on the same board, section by section, with narration, equations, diagrams, illustrations and live widgets you can drag.

The core is a learner model that updates from evidence. When you answer one of Y's checks, the answer is graded and labeled in the background, including which wrong belief it shows. At End session, Y updates a belief for each concept and skill, keeps a ledger of misconceptions, and writes teaching notes that must cite your real answers. The next time you reach that topic, Y teaches differently. It re-checks the mistake, says "Last time you...", and shows a chip explaining why it chose that step. You can switch Memory off and compare the same lesson without it.

You can see all of this in the "Mind" which is a 3D map of what Y believes about you, with a replay of how it changed and the evidence behind every neuron. Review brings back checks that are due. Trace shows which model did each step, with its timing and tokens.

Judges can try it without an account at https://ask-y-event.vercel.app. "Try as..." offers three demo learners in ML, computer vision and physics, each with weeks of history built through the same pipeline as real sessions and labeled as seeded. Each viewer gets a private copy, and new canvases grow that copy's Mind.

Y also tests changes to its own teaching. An offline GEPA evaluator proposes prompt changes and tests them on held-out simulated students. Its first candidate scored higher on validation (0.745 to 0.832) but lower on held-out (0.822 to 0.802), so the gate refused it and the live prompt did not change. We have not yet shown improved learning outcomes, and we do not claim to.

## More from Meta Global AI Developer Hackathon

- [jackpot-joe](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4671)
- [Consai](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4672)
- [Zoogom](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4673)
- [NorthTeak](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4675)
- [Fingertip](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4676)
- [Vroom Analytics](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4677)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4674. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt