NexusForge - Industrial Decision Twin

What is NexusForge? In manufacturing plants, a downstream motor overload alarm (RB-04) often triggers an unnecessary €2,450 servo replacement and 210 minutes of downtime, when the true root cause is an upstream €18.50 Viton pneumatic seal (PN-12) starving the line. Conventional dashboards stop at passive alerts, while standard chatbots hallucinate repairs without checking physical inventory. NexusForge changes the outcome. End-to-End Agentic Workflow (Shown in the Demo) 3D Digital Twin & Dual-Source Investigation (Evolus Agent #1): From the 3D assembly line twin, operators trigger an investigation that fuses live IoT sensor telemetry (pressure_bar, torque_pct, temp_c) with OEM technical manuals indexed in Evolus Document Extractors. Shannon Information Gain in the Counterfactual Lab: When multiple fault causes compete (51% pneumatic leak vs 33% regulator instability), NexusForge calculates Expected Information Gain (+0.624 bits) to prescribe the single quickest physical check for the operator. Once confirmed, posterior confidence updates dynamically from 51% to 91%. Inventory & Lead-Time Constraint Solver: Evaluates real-time spare-part stock (0 vs 15 units), supplier delivery lead times (180 min), mathematically blocking machine teardown if required parts are out of stock. Human-in-the-Loop & Evolus Workflow Dispatch (Evolus Agent #2): After the technician completes machine-side checks and signs off by name, a second Evolus Agent compiles the full analysis and triggers an automated Evolus Workflow via REST API, generating the complete maintenance activity package, root-cause diagnosis, and part requirements. Verified Impact & Tech Stack €11,439 saved per avoided misdiagnosis incident & -85% downtime (30m vs 210m). Built with: Open-weight LLMs on AMD Instinct GPUs (ROCm vLLM), Evolus AI (Agents, Workflows, Document Extractors, Knowledge Libraries), Bayesian Entropy Solver, Node.js, SQLite SHA-256 Audit Trail & Docker (52/52 automated tests passed).