Class Action Early-Warning System — Hearing the smoke alarm before the fire

The problem

For an automotive company, a class-action lawsuit is a nine-figure problem — and it rarely comes out of nowhere. It starts as a pattern: the same component, the same failure, described in slightly different words across hundreds of public complaints. Humans can’t read tens of thousands of complaint narratives looking for that pattern. So the first time legal hears about it is usually too late.

What I built

A system that reads public NHTSA complaints and related consumer feedback continuously, cleans and summarizes the text, and looks for the linguistic fingerprints of escalating defect patterns — clusters of complaints that historically preceded litigation. When a pattern crosses the threshold, legal and quality teams get an early flag while the issue is still fixable.

What changed

• –Risk detection speed improved by 25%
• –Legal and quality teams moved from reacting to lawsuits to investigating patterns before escalation

Under the hood

Python NLP pipeline: web scraping, text cleaning and summarization, n-gram pattern analysis, deep learning classifiers (LSTM/Transformer-based). Solo build.