# BenchMate

- **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:** [Stephen Sims](https://cerebralvalley.ai/u/simss4241)
- **GitHub:** https://github.com/simss4241-cmyk/BenchMate
- **Demo video:** https://youtu.be/q9skyzp6COc
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1084

BenchMate — Nothing is junk. Some things just need a plan.

I salvage electronics, and the hard part starts after taking something apart: remembering what I recovered, reading tiny markings, locating useful components on a board, and figuring out what I can build without counting the same part twice. BenchMate turns that growing pile into a workbench with memory.

Four connected spaces carry a find from discovery to a project:

• Stash: photograph parts, donor devices and tools; review candidate identities; keep quantities, condition, storage location and evidence together.
• Boardroom: map candidate components directly onto board photos with numbered, editable outlines. Keep board-specific knowledge in a Parts Cache, and record physical recovery before components become available stock.
• Build Ideas: explore builds with a bench partner in project-specific conversations. Photo uploads can prepare editable Stash cards; supported parts requests become validated application actions with saved receipts.
• Projects: review suggested requirements, match candidate parts from inventory, reserve available units, mark installations, return reusable parts, and track progress and next steps. Suggested requirements stay separate from reserved stock.

Muse supplies photo interpretation and planning; BenchMate supplies persistent records and transaction rules. A donor's possible contents never become recovered stock just because the model mentions them. Identification, user confirmation, working condition and compatibility remain distinct, and uncertain markings or missing evidence stay visible.

Built during the hackathon: the four-tab React/TypeScript workbench, Express/SQLite persistence, Meta API integration, editable photo and board proposals, scoped conversations, validated stock actions, recovery provenance, suggested-parts planning, and a separate browser-only judge preview.

The public source passes 76 automated tests. A fresh-checkout smoke test exercised real Muse label review, saving, reservation, restart/reload persistence and returns using isolated test inventory. The video shows the working app; Juno's Muse-agent cameos are scripted.

Explore the demo and interactive preview: https://neuroforged.net/lab/benchmate/
The preview uses sample data and recorded examples with no live AI, account or API key. The full app runs locally with the builder's own backend Meta API key. Automatic datasheet retrieval and electrical verification remain future work.

## More from Meta Global AI Developer Hackathon

- [GhostRep](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1081)
- [Lesstruct](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1082)
- [RestaurantOS AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1083)
- [Velohud](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1085)
- [Vouch](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1086)
- [OnboardIQ](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1087)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1084. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt