# Team AurorAI

- **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:** [Mikko Nykyri](https://cerebralvalley.ai/u/kkrat0s)
- **GitHub:** https://github.com/kkrat0s/aurorai
- **Demo video:** https://youtu.be/r_RPs_ZdsrU?feature=shared
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4511

AuroraSense turns an everyday smartphone into the starting point for a space-weather investigation.

Aurora maps, professional ground measurements, and a phone’s magnetic spike answer different questions. A nearby object or device movement can produce a dramatic reading; a modeled auroral oval does not confirm a sighting. AuroraSense helps learners, makers, and citizen scientists investigate those differences.

The working local prototype connects three views:

* Storm Track replays original NASA arrival predictions with submission times and revisions, making forecast disagreement visible.
* Aurora Map presents NOAA’s modeled activity with source-valid times and viewing context, keeping missing information explicit.
* Ground Truth supports consented foreground phone recording, stop/export, reproducible quality checks, and professional USGS reference context.

Meta Muse Spark 1.3 coordinates an investigation through seven typed scientific tools. It can inspect coverage, retrieve reference evidence, request validation, and propose an opt-in observation. Fixed authorization and scientific rules control which actions are permitted and what verdict the evidence supports.

The hackathon build includes the Next.js dashboard, Expo recorder, TypeScript quality checks and coordinator, Python evaluation, and reproducible evidence reports. NASA, NOAA, and USGS data are credited external inputs.

The evidence is inspectable: an owner-provided ten-minute iPhone recording delivered 5,985 samples and passed the unchanged acquisition checks. A separate real Spark run examined USGS evidence and produced an INCONCLUSIVE report when independent phone coverage was absent. A retrospective forecasting experiment reports its losses against simpler baselines.

The next experiment is synchronized phone and professional-reference measurements across independent sites. Regional phone-storm detection and phone-driven forecasting gains remain research questions. AuroraSense makes that investigation tangible, educational, and honest.

## More from Meta Global AI Developer Hackathon

- [Swapman](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4508)
- [Cognova](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4509)
- [ChatDoes Com](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4510)
- [JackLabs.ai](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4512)
- [Khadzhi Sarakaev](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4513)
- [Safecheck Ai](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4514)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4511. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt