Building Lexie: How I Used Google's Gemini Live API and Vertex AI to Transform Legal Intake

This post was created for the Google Gemini Live Agent Challenge. Follow along with #GeminiLiveAgentChallenge The Problem Here's something most people don't know about personal injury law: 39.5 million people get injured every year in the US , but only 400,000 actually file lawsuits . That's just 1%. What happens to the other 99%? They either handle claims themselves (getting 3.5x lower settlements on average) or get nothing at all. Why? Because small plaintiff law firms - the ones representing everyday injured people - are drowning in manual work. A single case requires: 90-minute intake interviews 40-60 hours of document review 8-12 hours of damages calculation At $2,350 in pre-settlement costs per case, firms can only afford to take cases with $100K+ settlement potential. The rest get turned away. Meanwhile, large defense firms (representing corporations and insurance companies) are getting cutting-edge AI tools. The justice gap isn't just wide - it's getting worse . So I built Lexi