# Verityn

- **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:** [Nomathemba Ncube](https://cerebralvalley.ai/u/commit2Noma)
- **GitHub:** https://github.com/commit-to-Noma/Verityn
- **Demo video:** https://youtu.be/MRwITo-nEOs
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2892

Verityn is an evidence-first STEM laboratory built around one idea: answers are everywhere, understanding is not. I'm a second-year CS student who realised I had been reading AI answers instead of learning, so I built a lab where thinking comes first.

Five beliefs shape every screen: thinking stays human, so learners predict before the AI speaks; practice should not need a lab bench, so they build, reset, and retry safely in the browser; diagrams should not stay static, so sketches become testable models; a wrong answer should teach something, so predictions are compared against measurements; and curiosity needs room to play, so the Experiment Bay brings back messing about on purpose.

Built for students who can't count on a lab bench or second chances, especially in Africa, where I study: everything runs in the browser, and every experiment can be reset and retried safely.

In the Guided Lab, learners predict (Yes / No / Not sure) before anything is tested. A deterministic modified-nodal-analysis solver in Python measures the circuit, so the result is physics, not a guess. A Muse Spark coach then climbs a bounded help ladder: a question, a pointer, and a demonstration only with the learner's consent (with Undo). In Solo, the server refuses every coach call, so the learner must apply the idea alone.

The Experiment Bay is the playground: wire anything, stack batteries, push the voltage with a stress rig until bulbs overwork and explode, then ask Muse Spark why, from the measurements. Reality Lens turns a photo, sketch or diagram into a reviewable circuit model: Muse Spark proposes the parts, the learner confirms, and the solver tests it. Spoken answers are transcribed by Muse Voice Transcribe.

Meta models propose; code decides. Model outputs are schema-checked, checked for leaking the answer, and checked against the solver's numbers, with a clearly labelled deterministic fallback. Circuit region detections shown are precomputed samples (Segment Anything Model 3.1, sample-only); no live SAM calls are made. Circuits are the first live module of four planned STEM disciplines.

Under the hood: a deterministic MNA solver; an evaluator returning CONSISTENT, CONTRADICTED or UNASSESSABLE; a Flask API with typed errors; mocked-HTTP backend and frontend suites with lint and build gates; and an end-to-end browser smoke test.

Verityn is a working prototype with no user data yet; classroom pilots are the planned next step.

Built solo for the Meta Global AI Developer Hackathon (Indie track, Build for Your World).

## More from Meta Global AI Developer Hackathon

- [Evangelina Hoepfer Tinkers Space](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2889)
- [Hot Seat](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2890)
- [Ghost Run](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2891)
- [Minerva X Tech](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2893)
- [SudoStudio](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2894)
- [Dylan Goldblatt](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2895)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2892. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt