Skip to Main Content

One Cube Up

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

One Cube Up — Demo video

MES Copilot for Small and Mid-Size Manufacturers Plant knowledge that today lives in operators' heads and free-text notes, turned into structured, searchable, costed production intelligence. Live: https://mes-copilot.onrender.com The problem: Small and mid-size manufacturers run on tribal knowledge. A stoppage gets scribbled as "line stopped, smelled hot." A supervisor who wants to know what Line 2 cost the plant this week waits for someone to dig through spreadsheets. Quality checks live in one experienced operator's eyes. Enterprise MES platforms solve this at enterprise prices and timelines, so plants under a few hundred people mostly go without. What we built: A working MES copilot over a synthetic, ISA-88/95-shaped plant dataset (Riverbend Fabrication, a fictional two-line metal-fabrication plant, 30 days of orders, downtime, inspections, and holds), running live on the Meta Model API. Three flows: 1. Vision QC: an operator photographs a stamped bracket; the vision model returns a verdict (pass/flag), defect types from a defined vocabulary, a confidence score, a plain-language note, and a disposition (use/rework/hold). Unrecognized photos are flagged for a human, never auto-passed. 2. Downtime Detective: an operator describes a stoppage in their own words; the model files it in plant format — line, equipment, class and cause code from a standard taxonomy, duration, root-cause and corrective-action drafts — costed at the line's downtime rate the moment it lands. 3. Ask Production + Shift Handoff: plain-English questions answered from figures computed in code (the model phrases, never calculates), with an honest "not covered" when the data cannot answer; one click produces a shift-handoff note. Supporting pieces: a Help chatbot that answers how-to questions from a built-in documentation knowledge base (retrieval in code, model phrases only what the docs say), and Plant Setup, where a user enters their own plant name, line names, downtime rates, and equipment so new filings are labeled and costed their way. Clean-room note: everything is synthetic and original, shaped only on public ISA-88/95 concepts. No employer or client code, data, or documents were used.

Team