AI For Good: Fighting Health Insurance with AI

Almost everyone in America has a insurance horror story, and at Fight Health Insurance we've built tools to help people fight back. In this talk you will learn how to use your tech skills to make the world a better place, evaluate the trade-offs between structured v.s. unstructured flows (which we call Doughnut, I might bring doughnuts – no guarantees), and details around privacy preserving AI. We’ll go deep on how we fine-tuned models for high-stakes denial language, built fallback systems to catch LLM flubs, and scaled our infra with vLLM, Ray, and sheer stubbornness. Along the way, you’ll see what works (and what crashes) when real users interact with AI in the wild—especially when they're scared, frustrated, or just trying to afford insulin. Whether you’re tuning models, designing interfaces, or wondering how to responsibly drop AI into a Kafkaesque system that still runs on PDFs and fax machines, this talk offers lessons from the frontlines—plus some surprisingly hopeful outcomes.