# strollcall

- **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:** [Zhuoyang Hu](https://cerebralvalley.ai/u/xiaohutongxue)
- **GitHub:** https://github.com/zh315-bit/strollcall.git
- **Demo video:** https://youtu.be/wXZCRLIrUCc
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2109

StrollCall gets you outside. Many of us want to go for a walk but don't know where to go, and "we should hang out sometime" with friends rarely turns into a plan. StrollCall turns one sentence into a short, real walk, on your own or with a friend.

How it works: say or type what you feel like, e.g. "25 minutes, quiet streets, coffee at the end", and pick a start point. StrollCall plans a 20–30 minute walk past places nearby, ending somewhere worth arriving. To walk with a friend, send one link: your friend taps "I'm in" on their phone, no app or account needed, and adds what they'd like. StrollCall finds a meeting point that is fair by walking time, not by distance on a map, and then a shared walk to a destination that fits both of you. In-app guidance follows you with a full-screen map, turn prompts and a voice walking buddy that can change the route mid-walk ("add a coffee stop", "I only have 10 minutes").

Meta models do the human side:
- Muse Glimmer (open weights, self-hosted on Modal) reads each wish into structured constraints.
- Muse Spark reconciles two people's preferences, explains the trade-off, ranks candidate plans and powers the walking buddy through tool calling.
- Muse Voice Transcribe turns speech into text; Muse Image illustrates invite cards.

A deterministic planner on our own New York walking graph (built from OpenStreetMap) handles geography in milliseconds. Model output is validated against real data: an automated check rejects any claim our data can't support, so we never invent places, opening hours or routes.

Every evaluation was pre-registered. On 12 New York test pairs the planner gave the correct outcome in 12/12, with a median arrival gap of 0.2 minutes. Meta's reading caught 10/10 pre-registered wishes vs 7/10 for keyword rules, and Muse Glimmer read wishes in a median 0.55 s vs 3.58 s for the hosted API.

## More from Meta Global AI Developer Hackathon

- [Brieblue](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2106)
- [NULLPTR](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2107)
- [Axis](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2108)
- [Fill Aloud](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2110)
- [Axiom Earthlab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2111)
- [Bartan Labs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2112)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2109. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt