Google Engineers Create Portable Offline AI Translator on Raspberry Pi 5

Google showcased the capabilities of one of its Gemma AI models on a local device in offline mode. Engineers built a fully autonomous translator based on a Raspberry Pi 5 single-board computer and the Gemma 4 E4B neural network running locally. The software was developed under the Google Gravity project, and the Gemma model powers the AI functions needed for translation between languages. The team also 3D-printed a compact case housing the electronic components, including a small screen for displaying translated text, a microphone, a speaker, a voice input activation button, and a language selector. The Gemma 4 E4B model runs on the device using the LiteRT-LM framework, and the Moonshine system handles text-to-speech conversion. Users can opt to have translated text displayed on the screen instead of spoken aloud. Once set up, the device works entirely offline, making it usable anywhere without an internet connection.