Fine-tuned civic grievance classifier

The brief

City help desks route thousands of complaints a day by hand. Fine-tune a small model that sorts complaints (roads, water, garbage, streetlights) by department, and compare it with prompting a large hosted model. Good looks like: a labelled set, accuracy per department, and a recommendation with cost and privacy.

What you will build

1. Prepare labelled complaints from public service-request data
2. Fine-tune a small model with LoRA
3. Prompt a large model on the same test set
4. Compare accuracy per department, speed and cost
5. Recommend one for the city

Free, open data to build it with

• NYC 311 service requests ↗ NYC Open Data

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.