When AI is wrong on a factory floor, the line stops. So we ground every answer in the ontology, hand over control only as far as it has been verified, and pursue autonomy only in factories that have been standardized.
A general chatbot produces plausible sentences, but a factory floor demands evidence, not plausibility. VEXPLOR retrieves across five paths at once to cross-check the evidence, and verifies the reasoning path itself as a graph.
Because an ontology already defines how items, equipment, processes and quality connect, an answer can only come from within those defined relationships.
FIG.3 — KNOWLEDGE GRAPHLinking items, equipment and processes is the starting point. The question a factory actually needs answered comes next — "what can stop us next week?" A schema holding only nouns has nowhere to ask it.
So we do not write a factory down as a parts list. People, equipment, materials, methods and IT systems can each be removed and the factory still stands — through outsourcing, a spare machine, an unmanned line, or paper. Every one of them is an interchangeable means. What remains after every substitution is what makes a factory a factory.
Nowhere in the five does the word "person" appear. Whether a human or a machine performs them is implementation, not definition. That is why an unmanned factory is possible — and why removing people does not change what a factory is.
AAF splits the AI's authority into five levels. The factory decides how far to allow, and commands that violate physical constraints are blocked by SHACL rules before execution. Its design references the governance requirements of regulated industries (IATF 16949, FDA, ISO 13485) — this is not a claim of certification.
Data viewing and status queries only
READ-ONLYSubmits its judgments as proposals only
PROPOSE · HUMAN DECIDESExecutes after operator approval
HUMAN-IN-THE-LOOPExecutes automatically within a verified scope
BOUNDED AUTONOMY · AUDIT LOGImmediate intervention on safety events
SAFETY OVERRIDEThe three core AIs — predictive maintenance, quality prediction and process optimization — built and applied as field projects.
SITES: a metal casting manufacturer · a materials company (IN PROGRESS)Expanded into an agent platform on top of the ontology. Six sites across Gen-1 and Gen-2 combined.
SITES: semiconductor parts maker · surface-treatment company · plating line · precision machinery parts maker (4 SELECTED 2026)We led the consortium developing the national manufacturing-data standardization reference model (AAS/IEC 63278), and the product design references the requirements of regulated industries such as IATF 16949, FDA and ISO 13485 (not a claim of certification). Equipment integration follows open standards like OPC-UA and MQTT.
20+ build sites, documented by sector and technical stack.