# Ripple handoff

- **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:** [Quentin TENNEREL](https://cerebralvalley.ai/u/SaladeChaude)
- **GitHub:** https://github.com/Quentintnrl/meta_hackaton
- **Demo video:** https://youtu.be/X1512R95KlY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3770

Ripple_handoff: a five-agent communication team for people who have never done communication.

For our final project, my university asks us to produce content, run a market analysis and write a brand strategy for a real product. Nobody in the team has been trained for that. I was curious whether the Meta models could do more than answer questions about it: whether they could do the job with us, the way a small agency would, while we stay the ones who decide. Ripple_handoff is that experiment.

It is a studio where five agents on the Meta Model API work as a team around one product, here Mairie360, a workspace for small French town halls that I build on the side.

- The Scout watches the field. It runs as an agent on the Responses API with two tools, search and open: it finds the comparable players, opens their sites and pages, reads their real rhythm, formats and tone, confirms event dates on the event pages. One real pass: 45 searches, 14 pages opened, every step kept and shown.
- The Analyst reads our own numbers, the real Instagram and LinkedIn posts, and says what to change first, in one line per fact, without repeating the chart.
- The Strategist proposes pieces of content, each with the situation it answers, who exactly it is for, its first line, the channel, and what we will count a week later to know it worked. It reads our refusals and what we already posted before proposing again.
- The Writer writes what we accepted: caption, hashtags, a carousel or a storyboard, pictures with Muse Image on demand.
- The Director plans the format mix and sums up each pass.

What makes it a team and not a chatbot: it works when we are not there. Put it on duty and it passes once a day on its own; the first screen the next morning says "the team's last pass, on its own, nobody asked", one line per agent. Every agent leaves a trace of what it read before calling the model, counted by the server, so we can check it. We only click yes, no, or rewrite. Nothing is written or posted without us, and a pass stops with one button.

The loop closes on reality: a posted piece is matched with the real post, and a week later the card shows what we promised to count against what it got. The Strategist reads that before proposing the next one. That is the part a market analysis course never gave us: a strategy that corrects itself on real numbers.

Under the hood: the Responses API in strict JSON with zod schemas on the server, the tool loop for the Scout, Muse Image for visuals, an abort signal through every call for the Stop button, every number computed server-side and never by the model. Node 24, Hono, SQLite, React. The demo dataset is a real pass on real posts, so the studio can be explored without spending a single call.

## More from Meta Global AI Developer Hackathon

- [Say Again](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3767)
- [SIkai Saathi](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3768)
- [FollowAlong](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3769)
- [BOLUS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3771)
- [Lisovate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3772)
- [Abra](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3773)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3770. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt