Case 06 of 07 · Transcritor, an in-house meeting transcription tool
Meetings become minutes without leaving the machine.
Sensitive meetings needed a written record. Local transcription separates speakers and produces minutes for review without sending the audio to the cloud.
AI-generated photo.
Scene · 17:05
What happens at 17:05
The meeting audio becomes text with speakers on the machine’s own GPU, and the minutes go out for review without the file ever touching the cloud.
Local transcriptionon this machine’s GPU
Processed100%
- Speaker 1 · 00:04Let’s start with the quarterly budget.
- Speaker 2 · 00:19September came in below forecast.
- Speaker 1 · 00:37Then the new hire waits until November.
Minutes for review
Decision: hiring moves to November. The audio never left the machine.
Illustrative scene, with synthetic data.
In production
What we built
Audio and video transcription on a local GPU, without the file ever leaving the machine
Speaker separation
Meeting minutes generated from the transcript
Results showing up in real time while the file is processed
Architectureat a glance
- Audio or video file
- Transcription on a local GPU
- Speaker separation
- Minutes and summary for review
Governance
Where a person decides
A person in control
- The generated minutes are a draft: whoever ran the meeting reviews them before they go out.
Risks handled
- Processing is 100% local: the audio never leaves the machine.
Technology · Whisper on CUDA, speaker separation, FastAPI, React
40-minute call, free
Want something like this?
On the 40-minute call, free of charge, we show you how a system like this would work in your operation.