I built the auto-ticket-categorizer repository on GitHub to triage incoming support tickets into Billing, Technical, HR, or General departments using a lightweight TF‑IDF + Logistic Regression pipeline, with confidence scoring and keyword‑based urgency tagging.
The dataset contains 80 manually labeled tickets, 20 per category.
I use a Logistic Regression model with regularization strength C=10.
Predictions below 60% confidence are routed to "Needs Human Review".
Urgency is tagged via keyword rules: urgent, down, not working, asap, crash, failed.
Evaluation uses 5‑fold stratified cross‑validation, achieving a mean accuracy of 0.70.