The Problem
Modern React and Next.js websites actively block synthetic JavaScript events, rendering traditional web-accessibility bots and extensions useless for users who rely on voice or assistive navigation.
The Solution
I've engineered a Hybrid architecture. A lightweight Chrome Extension captures 16kHz PCM audio, while a local FastAPI middleware routes the stream to the Gemini Live 2.5 API. When the AI executes a physical action, it connects directly to the active Chrome tab via the CDP debugging port (localhost:9222) to simulate raw, OS-level hardware clicks.
Architecture Diagram
