Selected work
Desktop · MLElectron · whisper.cpp · AWS

Transcriber Studio

Local + cloud transcription, end to end

An Electron transcription workstation blending on-device ML, wireless-mic hardware integration, and a serverless GPU cloud backend.

Role — Solo desktop + cloud engineer
  • Electron
  • React
  • TypeScript
  • whisper.cpp
  • WhisperX
  • AWS CDK / Lambda
  • GPT-4o
  • SQLite

Transcriber auto-imports from DJI wireless mics (fingerprinting transmitters by firmware filename), transcribes in real time with whisper.cpp and in batch with WhisperX on a local GPU or a serverless AWS backend, analyses with GPT-4o, and shares sessions over LAN via an HTTP + WebSocket bridge.

A task-token pull model keeps the cloud side cheap by avoiding idle GPU and NAT costs. Sessions persist to SQLite and export to PDF.

What's notable
  • Real-time whisper.cpp + batch WhisperX (local GPU or AWS)
  • Serverless AWS GPU backend with a task-token pull model
  • DJI wireless-mic auto-import via firmware fingerprinting
  • LAN sharing bridge, GPT-4o analysis, SQLite, PDF export