About this project
Every factory has the same fight: Production wants the machine running, Maintenance wants it fixed before it breaks. Today that gets settled in hallway arguments while risk keeps climbing. Downtime Arbiter is a multi-agent system that settles it properly. A Maintenance agent argues from real sensor and risk data. A Production agent argues from deadlines and cost but it only ever sees a coarse urgency level, never the raw numbers. That's what makes it a real negotiation instead of one model talking to itself. Maintenance's case isn't a guess. It uses causal reasoning based on a P-F curve, a known reliability engineering pattern for how failures escalate once a warning sign appears. So a 24-hour delay and a 72-hour delay can carry very different risk, and the system knows why. The two agents plan across a rolling two-week schedule, so one decision can affect what other machines are allowed to do too. When they can't agree, a rule-based Arbiter decides based on real cost, or escalates to a human if it's too risky to decide alone. Every step gets logged, so there's a full record of why each call was made. Built on the Nitrostack SDK, verified through Nitrostack's Test Cases, and deployed end-to-end on NitroCloud. External components: Zod for schema validation CWRU Bearing Data Center for the bearing-spall signal Groq + Qwen for agent reasoning.
Manufacturing & Industry 4.0 track
Create intelligent systems for smart factories, predictive maintenance, quality control, and supply chain optimization.