SecondMind

Video Demo

About this project

SecondMind is an ambient ai memory assistant built directly onto even realities g1 smart glasses. Humans subconsciously forget up to eighty percent of what they hear in daily lectures and conversations, and traditional note taking requires an active distraction that pulls you right out of the moment. We built this project to act as a seamless extension of your own memory by running a background audio processing pipeline that captures structures and privately feeds context right into your field of view without interrupting real world interactions. The entire system comprises over sixty one thousand lines of core code spanning c dart kotlin and typescript to balance low level hardware execution with deep cloud intelligence. The architecture completely decouples data synthesis into two distinct tracking pathways. When you wake the glasses with the phrase hey secondmind the system enters a completely silent tracking mode. This layer proxies raw audio streams from the foreground phone microphone through custom native c bindings for real time rnnoise reduction and voice activity detection. It then formats and streams text data as chunked packets over bluetooth low energy directly to the heads up display on the left and right arms of the glasses frame. When you need an active vocal sounding board instead waking the system with the phrase hey quacky triggers a conversational voice engine that streams natural low latency speech responses back to your ears using the elevenlabs api. When you sit down at your laptop the vercel hosted web command center provides full scale desktop memory synthesis. Because the mobile app constantly pushes data to a google firebase firestore synchronized memory store the web dashboard updates in real time. The intelligence layer leverages the gemini api to route tasks dynamically based on complexity. We use gemini three point one flash lite on mobile for sub second transcript parsing into typed models like notes and academic captures. On the we