đź› 2013 Smart Glasses Repurposed from Android to Linux as a Voice AI Assistant
The author revived the Recon Instruments Jet sports glasses: TI OMAP4430 chip, 1 GiB RAM, 8 GB eMMC, Kopin 428x240 microdisplay. Instead of Android, they installed Mainline Linux 6.18 with a boot time of less than 4 seconds and working drivers for the display, camera, microphones, Wi-Fi, Bluetooth, IMU, and GPS. The microWakeWord model runs locally on the glasses (using less than 2% CPU), while heavy inference—vision, ASR, LLM, and TTS—runs on a home server with an AMD Ryzen AI Max 395+ “Strix Halo” (128 GB) via llama-swap. Latency is masked by streaming, pre-warming, and pre-prepared voice responses. The author used Claude Code for reverse engineering and driver porting. The glasses feature scene description, chat assistant, a “world model” with object scanning and bounding boxes, OpenStreetMap maps, a shell, and even Doom via Chocolate Doom.
🌍 This case shows that LLM agents like Claude Code can genuinely take on low-level reverse engineering and driver porting for legacy hardware. The combination of local wake-word detection and remote inference on a consumer server provides a working voice AI companion without the cloud—a practical architecture for wearable devices, currently in the form of a single DIY experiment.
👤 A clear example of how “dead” 2013 hardware is transformed into a Linux platform with voice AI, a camera, and a world model. Important caveat: this is a single author’s story that has not yet been reproduced, and the drivers have not been released; the battery lasts about an hour at ~300mA, and Claude Code sometimes produces “spaghetti code” and makes diagnostic errors.
Source 1: https://blog.v0.net/converting-obsolete-sports-tech-into-ai-glasses-part-1/
Source 2: https://news.ycombinator.com/item?id=50499763