How the assistant works

Everything runs in your browser. No AI service, no API key, no data leaves the page.

1. Search the knowledge base. Your description is matched against 20 fix articles using BM25, the ranking formula behind most search engines. Rare, specific words count more than common ones.
2. Predict the category. A Naive Bayes classifier, trained on 224 labelled example tickets, estimates the most likely of 8 categories and how confident it is.
3. Security safety net. Clear security words (phishing, hacked, stolen…) override the model, because a missed security incident costs far more than a false alarm.
4. Set the priority. The ITIL impact × urgency matrix turns phrases like "whole office" and "can't work" into P1 to P4, with the same SLA targets as the service desk review.

How accurate is the category prediction?

Priority matrix