Local AI for checking form content

Collection in person and a request for a courier

In the order form a customer chooses „collection in person”, and in the notes writes „please send by courier”. Both parts are readable and complete, and the employee has to work out which to follow. The assistant shows the choice and the sentence from the notes side by side and prepares a question: „The form says collection in person, and the notes ask for courier delivery. Which way would you like to receive the order?”.

The whole sentence decides the meaning. The note „please do not send by courier” with collection in person contains no conflict, and a plain search for the word „courier” would raise a needless alarm. Every unnecessary question means extra work for the service team and another contact with the customer, so we count them separately from accurate flags.

What form field validation with a local model 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, 20% of today's work in the scenario. We measure both times in a trial on 20 to 100 cases the team has already handled.

What runs locally and what needs agreeing

We test the smaller model on the company's hardware and in the customers' language. When a result is not ready in time, the form stays visible to the employee and follows the usual review path. The administrator sets which part of the order goes to the app, where copies stay and what the logs record.

First remove the ambiguity from the form

If similar cases keep coming back, look at the form through the customer's eyes: is the collection choice clear, and does the notes field look like a place to change that choice? Ordinary field checks are still needed: MDN describes built-in checks for required fields and formats. When the dependency follows from clear choices, a rule in your current system is enough. A model is worth comparing where employees read different wordings of the same intention. A fixed answer layout makes the connection to the app easier: Ollama describes answers with an enforced structure.

Where to start

The first scope covers one form and a chosen dependency between fields. The service team points to the cases that come back for clarification, and the form owner judges whether the layout can be improved. 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.