CueRep

I built CueRep end to end as a macOS desktop app using Tauri, Rust, and React. It captures the rep's mic and the call audio as two separate streams, so it knows who is speaking without a diarization model. It transcribes in real time across 50+ languages, including Hindi and Hinglish. Keyword-matched coaching cards appear in under 2 seconds. When no card matches, a RAG pipeline answers from the company's knowledge base in under 5 seconds. Prompt caching cut the AI cost of those answers by about 90%. It works over Zoom, Google Meet, WhatsApp, or any other call app.