LaporPak!
- A web platform where citizens file public complaints and government staff triage, route, and resolve them. An AI pipeline classifies each report, scores its urgency, and assigns it to the right unit before anyone opens it manually.
About
LAPORPak! is a complaint management system built around three user roles: citizens who submit reports, admin staff who distribute tickets, and field coordinators who handle and close them. When a citizen submits a complaint, the system runs it through an AI pipeline before saving anything. A TF-IDF and Logistic Regression classifier (trained on NYC 311 open data) predicts the complaint category. A rule-based scorer then weighs the predicted category, the location, and the report text to produce a numeric priority score. Each ticket stores a plain-text explanation of that score so staff can see exactly why one complaint ranked above another. A duplicate detector using cosine similarity and geographic distance (Haversine) catches near-identical nearby reports and filters them from the consolidated view. Tickets that go past their SLA deadline are escalated automatically by a background scheduler, without any manual trigger. The stack is intentionally straightforward. FastAPI on the backend, SQLite for storage, scikit-learn for the AI components, and Vite with React and Tailwind CSS on the frontend. There is no message queue, no separate worker process, and no external AI API calls.