Support ticket to structured data

The brief

Customers describe problems in their own words, and engineers retype the details into the ticketing system. Use an LLM to turn each ticket into validated JSON (product, severity, steps to reproduce, environment), and measure how often it gets each field right. Good looks like: a schema, validation that catches bad output, and accuracy per field on at least 50 hand-labelled tickets.

What you will build

1. Define the JSON schema with Pydantic
2. Label 50 tickets by hand as a test set
3. Extract fields with structured outputs and validate them
4. Measure accuracy per field and cost per ticket
5. Handle failures and unsure cases with a review queue

Free, open data to build it with

• Customer support tickets ↗ Hugging Face

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.