# Ale + Jolly — Physical AI for Business

- **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:** [Jamilton Valkor](https://cerebralvalley.ai/u/JamilValkor)
- **GitHub:** https://github.com/JamiltonValkor/ale-jolly
- **Demo video:** https://www.youtube.com/watch?v=ygg_E0sYswc
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4199

Ale + Jolly gives a solo business owner an AI team with a body. It comes from a problem I know firsthand from my wife's ecommerce business: one person has to keep inventory, sales and marketing connected while running the day-to-day.

Jolly is a robot we designed and 3D-printed ourselves: a Raspberry Pi 5 with an expressive screen face, microphone, camera and servo arms. It sits on the owner's counter. Say "Jolly" and it wakes up, listens, answers, gestures, and goes back to sleep. Instead of opening another dashboard, the owner just talks to their business.

Behind Jolly is Ale, the business assistant inside SaleADS. One click deploys a dedicated cloud VM per business running Hermes, an autonomous agent runtime, with a team powered by Muse Spark: Ale coordinates an inventory specialist and marketing specialists. Using SaleADS MCP tools, they read business context, run scheduled jobs, propose strategies, and prepare ad creatives and landing pages, so work keeps moving when the owner is busy.

End to end: the owner asks Jolly to count incoming products. Ale asks for a photo; SAM 3.1 segments and numbers each item; Jolly shows the annotated image on its screen and says the count. The inventory agent receives the count with both images, and any stock change waits for the owner's approval. In parallel, the marketing team turns business context into proposals (a bet, creatives, a landing page) that the owner reviews in the web app.

Autonomy with the owner in control: the business "brain" records each idea with an explicit certainty level (data-backed fact, owner confirmation, inference or proposal), and the owner can replay how it evolved. Budget limits, approval cards and an activity trace keep decisions visible; launching campaigns and spending always require the owner's confirmation.

Brain in the cloud, body at home: on the Pi, small processes for voice, face, camera and arms talk over a local message bus. Muse Voice Transcribe streams speech recognition, and a paired, revocable device channel connects the robot to Ale, which sends back events that drive Jolly's face, voice and arms. The arms express interaction states and emotions through handcrafted gestures.

The submission includes a Docker Compose evaluation environment, a laptop robot interface and recorded interaction traces. Local cloud provisioning and SaleADS business services are clearly labeled simulators; live model calls use the evaluator's Meta API key. Muse Image provides labeled synthetic shelf scenes for reproducible tests, and a separate replay works without hardware or model credentials.

## More from Meta Global AI Developer Hackathon

- [Kaypro](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4196)
- [floodwatch](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4197)
- [Verity — Nnaemeka Eziamaka](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4198)
- [Voni](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4200)
- [TransitionBio](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4201)
- [ZKward](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4202)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4199. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt