# Spottr

- **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:** [Son Le](https://cerebralvalley.ai/u/son_le)
- **GitHub:** https://github.com/s0nle/Spottr
- **Demo video:** https://youtu.be/_9x-FlUKaeQ
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4312

An estimated 300,000 people drown per year worldwide according to the WHO. As a former lifeguard, I knew that this was a problem I've seen firsthand that I wanted to solve. Drownings in real life aren't anything like in the movies. There's no shouting, no waving, and people can slip under the surface in less than a minute. They can be quiet, unexpected, and deadly. In my experience, the hardest part of a lifeguard's job was keeping track of multiple swimmers in the pool, especially if you're fatigued from the hot sun or the freezing cold. Commercial drowning-detection systems exist, but they're built for large facilities that have the budget to afford custom cameras and setup. 

I built Spottr, an AI lifeguard assistant using Meta's SAM 3.1 and Muse Spark. SAM 3.1 was never built for live video. However, Modal GPUs allow Spottr to run the open-weight SAM 3.1 in real time. The camera's live RTSP feed runs through Modal GPUs running SAM 3.1, which finds every person's head and body in each frame. Spottr gives each person an ID, keeps that ID while they're underwater, and times how long their head has been underwater. Alerts are based off the 10/20 lifeguard rule, WATCH: at 10 seconds, ALARM: at 20 seconds, and EMERGENCY after that. The swimmer in distress is highlighted on the screen so lifeguards know exactly where to look. 

Before monitoring starts, the pool manager traces the pool's outline once, so Spottr knows when people climb in and out. People on the deck are also tracked separately from the people in the pool, and a live counter shows how many people are in the pool and how many are on deck. 

Muse Spark gives a second opinion on each alert at 10 sec as well as an end of session report. It can raise an alarm, but never cancel one. 

Spottr doesn't replace lifeguards; and no technology should. But it gives community pools, home pools, school pools, any pools that don't have the budget for large industrial systems an extra set of eyes, because Spottr never blinks.

## More from Meta Global AI Developer Hackathon

- [WatdOS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4309)
- [Brujula Libre SpA](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4310)
- [Equipa](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4311)
- [ZeroAndOne](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4313)
- [StateProof CrashLab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4314)
- [FairBuy](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4315)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4312. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt