Project Update: The Evolution of Gemini Live Vision (Omni-Guardian) We are thrilled to share the journey of Gemini Live Vision, our submission for the Google Gemini Developer Challenge. What started as a simple camera concept has evolved into a sophisticated, real-time multimodal safety assistant.
How it Evolved Phase 1: Sensory Integration We began by bridging the physical world with the digital mind. Using react-native-vision-camera, we established a stable 10-second snapshot stream that acts as the "eyes" of our agent.
Phase 2: The Neural Core We transitioned to Gemini 1.5 Flash (and recently testing Gemini 2.5 Flash Lite) to process visual data with ultra-low latency. Our backend server now acts as a high-speed relay, managing WebSockets and multimodal prompts simultaneously.
Phase 3: Localization & Accessibility To ensure global impact, we localized the entire interface into English. The "Guardian" persona now provides calm, clear instructions in real-time, helping users navigate potential dangers like smoke, sharp objects, or accidents.
Log in or sign up for Devpost to join the conversation.