# BlueSky

- **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:** [Diya Sharma](https://cerebralvalley.ai/u/DiSha1)
- **GitHub:** https://github.com/diya-18/phc-pulse
- **Demo video:** https://youtu.be/SbCEE8MAvTA
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4752

I'm from Chandigarh, India and if you've ever been to PGI here, you've seen the queues: families from Punjab, Haryana and Himachal who've travelled hours, often for care that should have been available much closer to home. A lot of that starts small, at a local health centre that didn't have the medicine that day. When I asked a chemist in Chandigarh where patients go when the government dispensary runs out, he didn't even pause: "They come to us." And when I asked if shops borrow stock from each other when they run short, he said, "This keeps happening all the time." The medicine usually exists somewhere nearby. It just isn't where it's needed, and nobody finds out in time.

That's what PHC Pulse is for. It's an early-warning system for medicine stock in Primary Health Centers across Punjab, Haryana and Himachal Pradesh. Staff report in under a minute by photographing their paper register, saying what they gave out, or snapping the shelf, and Meta's models turn that into stock entries in Hindi, Punjabi or English for them to check and save. The app then forecasts how many days each centre has left, warns the district officer before the shelf goes empty, and suggests moving medicine from a centre where it would expire to the one running short. The officer approves (both states have to, if it crosses a border), the sender dispatches, the receiver counts what arrived, and the alert closes by itself.

Officers can ask questions in plain Hindi or English, and state admins get a weekly plan from Muse Wonder working with Spark sub-agents. The rule I stuck to throughout is that AI reads, counts and suggests, but people decide. Forecasts and quantities come from plain code, and nothing touches the stock ledger until someone presses Save.

It runs on 1,211 real health centres mapped from OpenStreetMap, with clearly labelled demo stock data, and I load-tested it with 5,000 PHCs: a state dashboard still loads in 1.5 seconds. You can try it live at https://web-mu-six-60.vercel.app.

## More from Meta Global AI Developer Hackathon

- [Astravahn](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4749)
- [MaudeSignal](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4750)
- [UTN Argentina](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4751)
- [rmk](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4753)
- [Natasha Gubernov](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4754)
- [Loreg](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4755)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4752. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt