Brew: I Built a Real-Time Voice AI Drive-Thru Barista with Gemini Live API and Google ADK

TL;DR I built Brew — a real-time, voice-first AI ordering system for coffee shop drive-thrus. Customers talk to an AI barista through their microphone, and it takes their order through natural conversation. No buttons, no typing, just speech. The AI listens, understands complex orders with modifiers, handles interruptions, and updates a live on-screen menu and receipt as the conversation flows. GitHub: github.com/thilak15/Brew The Problem I Wanted to Solve Traditional drive-thru ordering is broken. Long wait times, order inaccuracies, and staffing challenges plague the industry. Human operators handle one car at a time, miscommunication leads to wrong orders, and during peak hours, lines stretch around the block. I wanted to see if a live voice AI agent could do this better — not a chatbot with text-to-speech bolted on, but a genuinely conversational agent that handles the full complexity of real ordering: sizes, modifiers, corrections, interruptions, and multi-item requests. What Brew