# LayoutLab

- **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:** [Muhammad Meidy Noor Al Barry](https://cerebralvalley.ai/u/Albary)
- **GitHub:** https://github.com/EndPx/layoutlab
- **Demo video:** https://youtu.be/webE5e5WNbo
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3940

LayoutLab helps warehouse planners test a physical change before making it on the warehouse floor.

A shorter route does not automatically mean faster order completion: workers may contend for an aisle, goods may wait at packing, or a different stage may become the bottleneck. LayoutLab makes those consequences visible and checkable in one interactive 3D workspace.

Run a baseline, move a rack or packing station, rotate a rack, or change an agent's initial parking position. Validate the candidate, run the same workload, and compare completed orders, completion time and cumulative waiting. Geometry edits carry their storage anchors, picking approaches and navigation obstacles with them. The original scenario and run remain available; applying a candidate is an explicit user decision.

Meta contributes a bounded experiment loop: propose an allowed change, run the deterministic simulator, inspect feedback, and recommend a candidate linked to its measured run. A recorded live Meta investigation demonstrates service-duration experiments. The physical-layout assistant has separately passed fixture-backed integration tests; we do not present those as live model evidence.

Our differentiator is the evidence behind the interface. Reports include inputs, constraints, engine versions, changes and results, and can be recomputed without another model call. Invalid, worse and non-improving alternatives remain visible. A separate storage-assignment benchmark preserves all 18 live Meta runs and compares them with a non-LLM heuristic.

The deployed sample, manual editor and recorded replay can be explored without an API key. Live inference is currently unavailable; the judge path uses recorded evidence and offline verification. Scenarios are synthetic, service times are declared assumptions, and results are modeled outcomes rather than measured warehouse savings.

Try it: https://layoutlab.endpx.cloud
Demo: https://youtu.be/webE5e5WNbo

## More from Meta Global AI Developer Hackathon

- [4xbench](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3937)
- [Knack](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3938)
- [WalkQuote-Solo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3939)
- [VERA, the Verified Exam and Revision Architect](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3941)
- [OmniMind](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3942)
- [Digikeys](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3943)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3940. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt