Skip to content
Daniel Fragoso

Real-time Meeting Copilot

Desktop meeting copilot for macOS that transcribes system audio and microphone in real time, identifies who is speaking, and keeps a live summary, key points and answers about the meeting, with an iOS companion app.

Architecture

Tauri 2 shell in Rust that launches sidecar processes. A Swift helper captures audio with ScreenCaptureKit and reads on-screen names with Vision OCR; a Python engine streams both channels to Deepgram and runs LLM workers (summary, speaker labels, insights, Q&A, titles) over an event bus. Speaker identity combines voiceprints, re-diarization and window snapshots. An Expo iOS companion gets an alert when you are addressed by name, shows a Live Activity and lets you ask the meeting questions through a WebSocket relay. Backend on AWS CDK (HTTP API with Lambda, relay, hosting and monitoring stacks), OAuth device-code sign-in and short-lived scoped transcription keys.

Highlights

  • Sole author, 200+ commits
  • Speaker identification from voiceprints and screen context
  • iOS companion with Live Activities and push alerts
  • Short-lived scoped keys, secrets in SSM
  • Phone PWA spin-off for following a meeting you can't hear well