# Triad Studios

- **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:** [Dillon griffey](https://cerebralvalley.ai/u/Wanheda)
- **GitHub:** https://github.com/TriadStudios/owl
- **Demo video:** https://youtu.be/cKChUFgaHl0
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1447

I work closely with Datto network engineers every day, and the same problem keeps coming up. A network closet is typically a blind spot. A tech walks in with a ticket, and nothing in that room tells them where these cables go or what lives on the other side of them. They trust the ticket, because that's the job. When the ticket is wrong, nobody finds out until something goes down.

That's exactly what happened to Google this September. During routine maintenance, a procedural error unplugged every fiber path in a cloud zone in 13 minutes and took it offline for about four hours. Nobody went rogue. The procedure itself was the problem. I've seen smaller versions of the same thing, where the team follows the procedure and the procedure is wrong. I wanted to build something that catches it before the cable moves, not after.

That's OWL. When a work order is filed, OWL checks it against the live switch and flags it if the device isn't where the ticket says. In the closet, a Reachy Mini robot uses SAM to see the tech's hands on the switch. If they reach for the wrong port, it warns them before they pull. When the right cable comes out, OWL confirms it on the switch, and the ticket closes with a full timeline and a clip of each moment. The tech sees the real ports on their phone, or hands-free on Meta Ray-Ban Display glasses.

I built it the way the people I work with would want it. Safety is OWL's number one priority, so AI never makes the decision. OWL only reads the switch and never changes config. Verdicts come from a plain rules file. Muse Spark only phrases what Reachy says, and every line is checked against policy before it's spoken. And it's a platform, not a one-off: any ticketing system can plug into OWL's API and fit into the tools teams already use.

## More from Meta Global AI Developer Hackathon

- [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)
- [aria](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1448)
- [RanveerMeel](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1449)
- [Assent](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1450)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1447. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt