How goncalvestalie/poc-ia-transcription works
This Windows desktop proof of concept captures microphone audio, streams it to Voxtral for live transcription, and lets the user generate a structured Mistral summary with an editable prompt and copyable output. The runtime is concentrated in poc.py; the graph separates its UI, transcription, and summarization responsibilities. The README documents the overall audio-to-cloud workflow; excerpts show UI transcript handling and the summary API call, but omitted code limits verification of some internal wiring.
Main entry points: poc.py Summary request
Components 13
Desktop interface
- Desktop app
poc.py - Transcript view
poc.py - Prompt editor
poc.py - Summary output
poc.py - System clipboard
Live transcription
- Microphone
- PCM capture
poc.py - Voxtral realtime STT
- Live transcript
poc.py
Summary generation
- Summary request
poc.py - Mistral cloud API
- Structured summary
poc.py
Other components
- Desktop user
Connections 13
- Desktop user → Desktop app: operates
README.md - Microphone → PCM capture: provides audio
README.md - PCM capture → Voxtral realtime STT: streams chunks
README.md - Voxtral realtime STT → Live transcript: transcribes
README.md - Live transcript → Transcript view: displays deltas
poc.py - Desktop user → Prompt editor: edits prompt
README.md - Desktop user → Summary request: requests summary
README.md - Live transcript → Summary request: injects text
README.md - Prompt editor → Summary request: supplies prompt
poc.py - Summary request → Mistral cloud API: requests completion
poc.py - Mistral cloud API → Structured summary: returns summary
README.md - Structured summary → Summary output: shows result
README.md - Desktop user → Summary output: copies summary
README.md