# Metaladon

- **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:** [ROHIT MADHAVAN](https://cerebralvalley.ai/u/6easx)
- **GitHub:** https://github.com/ROHITCRAFTSYT/readback
- **Demo video:** https://youtu.be/JUVaLpOMgj8?si=yt5eBEVujclW_YaQ
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1443

Readback catches readback errors in air traffic control radio. When a controller says “descend flight level one hundred” and the pilot reads back “one one zero”, nobody may notice, and missed readback errors are a classic cause of level busts and loss of separation. Today a unit finds them only by listening to recordings by hand.

Readback listens to the frequency, parses every clearance, and runs two independent checks: the pilot’s readback against the controller’s instruction, and the aircraft’s ADS-B track against what it was cleared to do. Alerts appear on a live radar monitor. A review console lets safety teams upload up to 4 hours of recorder audio, confirm or reject each alert next to the audio, and export an alert register with a precision measured on their own radio. For any alert, a Muse Spark agent gathers the evidence through tool calls and drafts an occurrence report.

In simulation it caught 2,052 of 2,052 planted problems with no false error alerts. On a blind seven-airport test of real radio, a fine-tuned Muse Glimmer 30B parsed 86.3% of clearances exactly, the best of four parsers. Every alert goes to a human: it is decision support, not a replacement for the controller.

## More from Meta Global AI Developer Hackathon

- [RemnantX](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1440)
- [Showhow](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1441)
- [AINGEL — Independent Developer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1442)
- [Shoestring Studios](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1444)
- [Fahim](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1445)
- [VerdantAI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1446)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1443. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt
