On-device manufacturing AI
  • On-device AI — drawings never leave the shop
  • Direct control of 9 CAD·CAM systems
  • Ontology-driven deterministic reasoning

DION AI

02 · DION

Four letters, four technologies

  • Design — reads machining features from drawings
  • Intelligence — learns shop-floor know-how
  • Ontology — reasons within the knowledge graph
  • Network — connects 9 CAD·CAM systems
03 · Machine data

Machines, understood as data

  • Lathe, 3-axis, 5-axis — one ontology for all
  • Axis layout, travel and interference as relations
  • KBM applies machine limits to process decisions
04 · Chuck & jaw data

Gripping conditions, computed

  • Chuck and jaw strokes captured as data
  • Soft-jaw steps registered as instances
  • Grip feasibility judged per stock diameter
05 · Tool data

Tool selection with evidence

  • Tool and holder standards unified in the ontology
  • Feature → process → tool matched by KBM
  • Built on 5,061 machining knowledge mappings
06 · NC output

Verified NC code, delivered

  • FANUC 9 series · 11 machine-model posts
  • KBM cutting conditions applied with evidence
  • Output only after human approval
07 · In-house LLM

An LLM fine-tuned on the shop floor

  • Fine-tuned for CNC machining
  • NVIDIA Innovation Lab · 8×H100 · 60 days
  • Served on-device — no drawings leave the shop
08 · Partners

An ecosystem built together

  • Siemens technology partner · Hexagon Korea partner
  • NVIDIA Inception member · NVIDIA Innovation Lab
  • AWS and Lambda startup programs
09 · Graph

Design the process as a graph

  • Node-based process pipeline editing
  • Model import → features → KBM → NC
  • Human approval gate before execution
10 · Board

Instruct in chat, execute on approval

  • Natural-language direction → step proposals
  • Human approval before execution
  • Cutting conditions returned with evidence