01 Opportunity
The knowledge that keeps a production machine running rarely lives in one place. Here it was spread across electrical drawings in one folder, mechanical drawings in another, OEM manuals on a shelf, site SOPs in a binder — and the machine logic itself locked inside PLC code that only an engineer could read. When something odd happened mid-shift, the quality of the answer depended on who happened to be rostered on.
A major food manufacturer wanted that knowledge available to everyone on the floor, all the time — with one hard constraint. Drawings, programs and procedures are production IP; none of it could go to a cloud service. Whatever answered the questions had to live inside the fence and work without going online.
02 Solution
We built a digital twin of the machine for AI: the electrical drawings, mechanical drawings, OEM manuals and SOPs were ingested into one knowledge base, alongside the machine’s PLC program. The AI doesn’t guess at how a generic machine of that type usually works — it knows this machine: its circuits, its mechanics, its procedures and its code.
The whole system runs on an on-premises server on the plant network. No internet connection, no cloud subscription, no data leaving the site — it works completely offline, which is exactly how the production network is run anyway.
Operators and maintenance ask questions the way they would ask a colleague. The assistant explains the logic behind the machine’s code in plain English — why an interlock is holding a sequence, what conditions a step is waiting on, what the machine expects before it will start — translated from the program rather than guessed, and grounded in the drawings and manuals that describe the hardware around it.
03 Outcome
Answers that used to depend on the right person being on shift now come from a screen on the plant network, around the clock. Operators understand why the machine stopped instead of just that it stopped, maintenance traces faults through the logic faster, and new starters learn the machine by asking it questions.
Just as importantly, the knowledge now outlives any one person. Decades of machine know-how — spread across drawings, manuals, procedures and code — is captured in a system the whole crew can query, and not a byte of it ever left the site.
04 Project highlights
- Digital twin of the machine built for AI — electrical drawings, mechanical drawings, OEM manuals and SOPs ingested into one knowledge base
- Works without going online: runs on an on-premises server, so drawings, programs and procedures never leave the site
- Explains the logic behind the machine’s code to operators and maintenance in plain English
- Answers are grounded in the machine’s own documents and program — not generic internet knowledge
- New starters get the same answers as the veteran who has run the machine for years
- Available around the clock on the plant network — no waiting for the one person who knows the machine