Josue Arango
All projects

2026 · AI & Agents

IT Ticket Triage with Machine Learning

Python · scikit-learn · TF-IDF · logistic regression · live demo

A text classifier that routes IT service-desk tickets, written in English or Spanish, to one of seven queues: network, email and Microsoft 365, accounts and access, hardware, security incidents, printing and software. It is trained with scikit-learn, exported as plain JSON and scored in the browser, so the live demo needs no server and nothing typed leaves the page.

Technologies

Pythonscikit-learnTF-IDFLogistic regressionNumPyJavaScriptModel evaluation

My role

Sole author. Built the bilingual dataset generator, the training and evaluation pipeline, the JSON export, the JavaScript scorer with its parity test, and the demo page.

Highlights

  • Evaluated on 924 tickets written from templates the model never saw in training: 83% accuracy and macro-F1
  • Per-queue precision, recall and a confusion matrix published next to the demo
  • Explains each decision with the words that pushed the ticket into its queue
  • The JavaScript scorer matches scikit-learn's probabilities on held-out tickets (parity test)
  • Small enough to ship as a static file: 1,500 features, about 105 KB