The problem
Support teams plan staffing around ticket volume, but volume on its own does not tell you how much work is actually arriving. A hundred password resets and a hundred escalations look identical on a dashboard and consume completely different amounts of time.
The concept
If complexity can be predicted from the ticket itself rather than judged after the fact, resourcing decisions can be made ahead of the queue instead of in reaction to it.
What I built
A Flask application that ingests help desk ticket data, processes it at scale, and predicts the complexity of incoming tickets. It surfaces the bottlenecks that would otherwise only be visible once a queue had already backed up.
Technical implementation
Python and Flask for the application layer, with a machine learning model trained over a dataset of more than 100,000 ticket records. The bulk of the work was in the data processing pipeline: getting real help desk data into a shape a model could learn anything useful from.
What it taught me
Working support tickets every day is what made this project obvious to build. The pattern it looks for is one I had already been noticing manually in Intercom.