# Tokiarivelo

- **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:** [Tokiarivelo ANDRIAMANDROSO](https://cerebralvalley.ai/u/Tokiarivelo)
- **GitHub:** https://github.com/Tokiarivelo/mitsinjo-grid
- **Demo video:** https://youtu.be/GSu96kD7HRw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1951

Mitsinjo Grid is an AI copilot for a simulated rural solar microgrid in Madagascar. One microgrid powers a clinic, a school, water and every home. When the sun drops, the operator has to decide who keeps the light.

The operator builds a simulated week (sun, clouds, storms, random incidents in chaos mode) and sets the levers: dispatch strategy, clinic lifeline floor, night battery reserve, sector priorities and caps. The whole week plays in seconds across three views: a 3D village through day and night, a power topology from the solar hub to each sector, and a circuit view with every breaker. The Strategy Lab compares strategies on the same event. For example, with 70% less sun for 10 hours, proportional sharing gives the clinic 44% of its energy and essential-first raises it to 58%.

The operator can ask the copilot by voice ("Who loses power first?"). Muse Spark answers by calling the engine's tools, and every number in the answer links to the simulation run that produced it. In autonomous mode, the engine picks and applies each action inside the village rules: it shifts flexible loads, keeps the night battery reserve and isolates a sector on fire. Muse Spark explains every decision, and one tap takes back control.

The operator can also bring real objects into the digital twin. From a photo, Muse Spark names the buildings and appliances and SAM 3.1 outlines them. Muse Image draws a clean front elevation of a building. Muse Spark builds its 3D model in stages and scores each stage against the drawing. The operator then sets the object's load, places it on a feeder line, and every strategy is recomputed with its consumption.

Key rule: numbers come only from a deterministic TypeScript engine. The AI explains, proposes and drives, but never invents an energy value. All data is simulated. The app runs in English, French and Malagasy, in dark and light themes, on mobile and desktop.

Built solo from scratch during the hackathon (first commit Oct 26, 2026).

Live app: https://mitsinjo-grid.vercel.app/simulation

## More from Meta Global AI Developer Hackathon

- [Reflection Lab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1948)
- [ShiftSense](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1949)
- [The U-Team](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1950)
- [Pranav Paharia](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1952)
- [MindFlayer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1953)
- [Zukki](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1954)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1951. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt