AI for industry
AI and automation for Swiss manufacturing companies
Manufacturing generates more data than is ever analysed: stoppages, changeover times, quality measurements, fault messages. At the same time, a great deal of knowledge sits in the heads of a few people. We connect language models to controllers, MES and documentation so that shift supervisors, maintenance and management get answers in minutes rather than days.
Where AI makes a concrete difference in the plant
- Downtime and OEE
- “Why did line 3 stop last week, and which cause is costing us the most time?” Answers from MES and controller data instead of from spreadsheets that someone first has to copy together.
- Shift and production reports
- Reports are generated automatically from the shift's data and only need to be checked and supplemented by the shift supervisor.
- Knowledge from manuals and faults
- Operating instructions, maintenance logs and resolved faults become queryable. New staff find the solution that a colleague already found two years ago.
- Order processing in the office
- Preparing enquiries, quotes and purchase orders, bringing together delivery dates from ERP and production. Decisions and dispatch remain with people.
IT/OT from practice, not from the textbook
Schady Hamid works as an IT/OT project manager at an industrial company in the canton of Aargau and teaches at the ABB Technikerschule in Baden. The connections are therefore built from the plant's perspective: network segmentation, controller cycle times and the question of who gets called at night when something is not running.
By default, data is read via OPC UA or the existing MES interfaces. Write operations are only possible via a list of permitted operations, with approval by a human and a complete audit log.
Data stays in the plant when it has to
Much production data should not leave the plant. In that case, the MCP server and language model run locally, on your own NVIDIA hardware in your own network. Where a cloud platform is permissible, we use enterprise offerings that do not use inputs for training.
Limits we state up front
- No intervention in safety-related functions such as emergency stop or protective devices.
- No write access to controllers without an allowlist, human approval and an audit log.
- No queries that affect the controller's cycle times.
- No cloud requirement: fully local operation if necessary.
Frequently asked questions
Which controllers and systems does Kapitel H work with?
With OPC UA as the preferred interface, controllers such as Siemens S7, Beckhoff and B&R, as well as MES and control systems. What matters is that the data can be read via a stable interface. A feasibility study of one to two weeks checks this.
Do we need a cloud for this?
No. The MCP server and language model can run entirely in your own network, on your own NVIDIA hardware. A cloud platform is an option if the data allows it, not a prerequisite.
What happens if the AI suggests something wrong?
Write actions only run via a list of permitted operations and require approval by a human. Every action is logged. Analyses state their data source so that they can be verified.
Can our own people run it themselves later?
That is the goal. Maintenance and IT are trained, and the source code and documentation in German or English are handed over to you. On request, Kapitel H takes over operation and further development under an annual contract.
Let’s check whether it holds up for your systems.
The feasibility study takes one to two weeks and ends with a document you can keep working with even without us. If an interface cannot meet your requirement, it says so.
Request a first call