Documents that make a good knowledge base
Operating and maintenance instructions, service handbooks, wiring diagrams with descriptive text, spare parts lists, technical bulletins and documented fault reports all work well. The more structured the material, the more precise the answers. Older scanned documents usually need text recognition first.
Version control matters. Machines evolve over many product generations, and an answer has to match the right model and variant. We therefore tag every document by machine type and configuration.
Fault reports that record cause and fix are especially valuable. They hold field experience that never makes it into a manual, such as the usual root cause behind a particular symptom on a given model.
Safety first: what the assistant should not do
Language models can be wrong. Safety notices, lockout procedures and inspection requirements from the operating manual always take precedence. The assistant shows the original passage so the engineer can read it directly, and it points back to the source document on safety-related topics.
We also address know-how protection early on. Which documents may external service partners see, and which are for your own team only? Roles and permissions are configured in kalAIDoCreate.
For equipment with hazardous zones, it helps to flag safety-critical chapters explicitly. The assistant can then refer the engineer to the original instructions rather than paraphrasing procedures in its own words. The same applies to service back-office staff doing first-line diagnosis by phone.
Getting started
If the assistant later needs to talk to your ERP, ticketing system or machine data, our AI code engineers deliver that through neooutput, for example as an MCP server.
The pilot quickly shows which questions are answered well and where documents are missing or unclear. Closing those gaps improves your technical documentation as a side effect.
- In a free 15-minute intro call we pick one machine series with high service volume.
- You provide the relevant manuals and a selection of resolved fault cases.
- We set up kalAIDoCreate and test it with experienced engineers against real questions.
- After the pilot we extend coverage to more product lines or service partners.
Frequently asked questions
Does it work without internet at the customer site?
A language model normally needs a connection to its server. For sites with poor coverage, key documents can additionally be made available offline in a mobile app. We define the right architecture during the project.
Can it read error codes straight from the machine?
Not out of the box. Connecting machine data is possible as an extension, for example through an API or an MCP server.
Who maintains the content?
Usually technical writing or the service back office. New manual versions and resolved cases are added to the knowledge base, and outdated versions are removed.