What they built

NeuroTwin Evolver — a self-evolving industrial meta-agent that rewrites its own logic when the real world changes

NeuroTwin Evolver is a self-evolving industrial meta-agent built on the NitroStack MCP framework, designed for Industry 4.0/5.0 environments where a fixed automation policy breaks down the moment reality diverges from its training — a heatwave draining UAV batteries faster than expected, a sensor going noisy mid-flight, a carbon-tax policy shifting overnight. Rather than one model wearing every hat, the system is a genuine multi-agent pipeline: a Resource Manager Agent computes a fleet-wide directive by treating energy price, carbon intensity, and delivery pressure as one problem; a Logic-Refactor Agent proposes candidate operational-logic mutations for the affected unit under that directive; a NeuroTwin Validator Agent checks every proposal against symbolic safety rules (battery floor, safety margin, latency ceiling) before anything is allowed near the physical fleet; and a Protocol-Evolver Agent independently monitors swarm-wide communication health, re-engineering the network topology between leader-follower, mesh, and cellular structures when links degrade. The result is a fleet of UAVs, AGVs, conveyor cells, battery racks, and safety gates that detects, proposes, validates, and deploys its own fixes live — visualized through a real-time digital-twin map, a mutation-lab dashboard showing the full agent trail, and per-unit telemetry, all runnable and testable through NitroStudio.