MEND - X — Industrial Intelligence Platform
Edge-to-cloud industrial diagnostics and deterministic RAG platform delivering sub-second fault triage, lockout/tagout protocols, and OEM manual citations for plant technicians.
Performance & Business Metrics
Project Overview
MEND - X (Maintenance, Engineering & Networked Diagnostics — Next Generation) is an enterprise-grade industrial diagnostics and Retrieval-Augmented Generation (RAG) platform engineered by Dimensity Labs for manufacturing facilities, robotic workcells, and automated assembly lines. It implements a Three-Tier Intelligence Architecture (NORD edge tier, FORGE mid tier, APEX enterprise tier) to balance real-time edge latency (<350ms) and multi-manual reasoning. The system features a deterministic hallucination barrier with a hard-stop safety gate (>=0.75 cosine similarity baseline), dual-pass PDF ingestion (PyMuPDF + pdfplumber) preserving tabular schematics, dual-store persistence (PostgreSQL 16 with pgvector + ChromaDB + Redis), and an omni-channel technician suite spanning Next.js 16 and an Expo React Native mobile scanner.
The Problem
In manufacturing facilities, unplanned machine stoppages cost up to $250,000 per hour. When a critical alarm or fault code occurs, maintenance technicians typically spend 45+ minutes thumbing through 500-to-1,200-page OEM manuals or fragmented documentation while downtime cascades across production lines.
Our Solution
Engineered an edge-to-cloud 5-stage RAG pipeline (Classification ➔ Hybrid Retrieval ➔ Cross-Encoder Reranking ➔ Deterministic Gate ➔ Tiered LLM Generator) backed by FastEmbed (BAAI/bge-small-en-v1.5) and dual stores (pgvector & ChromaDB). Introduced a strict >=0.75 cosine similarity evidence validator that aborts low-confidence queries to prevent hallucinations, cross-manual entity disambiguation for collision-free fault lookups, and dual-pass extraction preserving pinout tables.
Results & Business Impact
Reduced critical downtime triage from 45+ minutes to under 350ms on air-gapped edge devices, guaranteed 100% deterministic source traceability with exact page/section citations, eliminated cross-manual fault collision, and delivered zero-hallucination safety protocols.