# AutoLab

- **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:** [Vincent Ricardo AMOUSSOU](https://cerebralvalley.ai/u/ric-amo)
- **GitHub:** https://github.com/hkdebendo/autolab
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1298

AutoLab is an autonomous AI lab for materials discovery.

Every battery, chip and solar panel starts with a material, and finding a new one can take years of trial and error. AutoLab hands that search to an autonomous AI scientist, built on Meta models.

A researcher states a goal in plain language, by text or by voice. AutoLab then runs the whole investigation in a closed loop:

Propose: new candidate crystals are designed by chemical substitution in known materials, each with a hypothesis.
Simulate: each crystal is built and relaxed atom by atom, and its energy is compared with every known phase of its chemical system.
Test: candidates go from the cheapest filters to the most demanding ones: charge balance, stability, electronic insulation, contact with the electrodes, and finally ion motion by molecular dynamics.
Learn: when a candidate fails, the code measures why (for example, "chlorine diffuses at 17.7% of lithium"), the agent turns the measurement into a hypothesis, and the next round corrects course.
Conclude: the finalists are ranked by explicit rules, checked against the literature, and the champion's band gap is verified by DFT. A scientific report, exportable as PDF, closes the research, followed by a review of the whole campaign and the next one to run.
Everything can be explored: the agent at work in a live lab view, every crystal in 3D, ions hopping through the framework next to the parent crystal, and a Results page with each champion's properties and error bars. The researcher can question the agent about its own work.

AutoLab is built to be trusted: thresholds are explicit, methods are checked against reference calculations, and blind benchmarks test the whole pipeline.

It is applied today to three battery challenges: lithium solid electrolytes, cobalt-free cathodes and sodium batteries. Its design extends to materials for any industry. The project is open source and its demo runs without a GPU.

## More from Meta Global AI Developer Hackathon

- [Dominik Szopa](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1295)
- [Team A.R.G.U.S](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1296)
- [sakanfar](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1297)
- [Team Bliss](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1299)
- [Team Eco cycle](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1300)
- [HaxyScott Technologies](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1301)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1298. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt