Local AI for document review before a process change

One change, references in many documents

An engineer prepares a change and looks for the instructions and forms that may need updating. Searching by file name is not enough when the references to a fixture sit in the text. An example: on a packing line the label of a fixture changes. The work instruction and the inspection form still carry the old label, and the company requires these documents to be processed in its own environment.

The assistant finds both passages and shows them with the line name, the version and a link to the source. The engineer opens the documents, judges whether the reference needs changing and records the finding at the place found. Until the change is approved, the existing documentation applies. The app does not publish new instructions and does not decide the technical scope of the change.

What local AI for manufacturing documentation costs and when it pays back

The count covers the time spent searching, reading and compiling the material. What stays is the time to check the output against the source, 25% of today's work in the scenario. We measure both times in a trial on 20 to 100 cases the team has already handled. With 40 changes a month at 90 minutes of searching for references, the count covers one person running changes in a mid-sized plant.

What runs locally and what needs agreeing

The model runs on a named computer or server in the company. What remains to agree is how documents are read, where the material prepared for search is stored and where findings are recorded. Where the model runs does not describe the whole path of a file through the app. A trial on the target hardware covers several people working at once and the time from a question to a checked source. According to Bartanowicz and Jassem (LREC 2026), on 146 questions about technical documentation in Polish, Bielik-11B led among the models compared. The choice of model is still settled by a trial on your own documents.

When your current register of links is enough

If the document management or product data system already links the fixture to its instructions and forms, those links are a good starting point. A more convenient review view or completing the register may be enough. AI is worth testing when the references sit in free text and the register does not lead to those places. Finding the approved version is covered in the article on local AI for formulation versions.

Where to start

The first scope covers one line and an agreed set of instructions and forms. For the call, describe the last change where related documents were hard to find. We roll out local models as part of private LLMs, and we build the view for working with sources as an AI app. For the free process review, a description of one case and the places where the material sits is enough. Confidential files stay in the company, and you get a written plan within 2 working days.