The Overview
React Native/Expo app overlaying real-time translated subtitles on live video calls. Audio streams through Deepgram for STT, translates via GPT-4o-mini, renders subtitles — all sub-second.
The Build
Streaming (not batch) STT, lightweight translation model (4o-mini), immediate render without waiting for complete sentences. Architecture prioritises latency above all else.
The Result
Working real-time translation overlay maintaining sub-second latency. Live audio in one language becomes readable subtitles in another before the speaker finishes their thought.