# Shunfeng

- **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:** [Hung Pin Hsiao](https://cerebralvalley.ai/u/fuluboy)
- **GitHub:** https://github.com/fuluboy/shunfeng
- **Demo video:** https://youtu.be/yWcVkrgqqIE
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1314

Shunfeng is an iPhone app for deaf and hard-of-hearing travelers. It listens to airport announcements and buzzes only when one is about your flight.

At the gate, gate changes, delays and final calls are spoken out loud, often in Mandarin first. If you can't hear them, you watch strangers: when they stand up, you stand up. Boards help, but nobody watches a board every minute.

How it works: you photograph your boarding pass (or type the flight number), press Start, and put the phone away. The microphone streams to Muse Voice Transcribe; simple rules on the phone pick out the flight number, the event (gate change, delay, boarding, final call, cancellation, a passenger being paged), the gate and the time. Only announcements for your flight reach you, as a vibration pattern you can feel in your pocket: one buzz for "maybe, please check", two for a change, three for "go now", one long buzz for a cancellation. Final calls repeat every 30 seconds until you tap "I'm at the gate". Announcements in another language are translated into yours, and the whole app works in English or Traditional Chinese.

It is built to be honest when it isn't sure. If it hears your flight number but not the news, or a different destination, it says "maybe" instead of sounding certain, and a large Chinese/English card helps you ask staff. For Taoyuan flights it also checks the airport's live flight data. It pauses during phone calls and without signal, warns you if it stops listening, and stops by itself when your flight departs.

Results: 47 of 47 alert decisions right on test announcements after speech recognition, 10 of 10 on a held-out set, 12 of 12 announcements caught live by the iPhone's own microphone 1–2 m from a speaker, all 8 fields read correctly on 12 of 12 real boarding passes (about 5 s each), about 5% battery per hour with the screen off, and roughly US$0.50 of API cost per trip.

The name comes from Shunfeng'er (順風耳), the all-hearing god who stands beside Mazu and watches over travelers.

Before the hackathon I had only written synthetic test announcements (included in the repo, clearly labeled, with the organizers' permission), tried parsing them in Python, and saved snapshots of Taoyuan's open flight data. All app code in the repo was written during the hackathon.

## More from Meta Global AI Developer Hackathon

- [Crucible](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1311)
- [FieldProof by 180Climate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1312)
- [OPIE CHISOM SAMUEL](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1313)
- [Omni-Ouro Inc.](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1315)
- [PAVLENS AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1316)
- [Deep Shah](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1317)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1314. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt