Transcription & Translation

Self-Hosted Transcription

PanOps uses a state-of-the-art open-weight transcription model, running on PanOps-operated GPU infrastructure. This is a critical design decision: your audio recordings never leave your secure storage to be processed by a third-party API. The PanOps transcription engine runs inside the same infrastructure boundary as your data, and transcripts are written directly to your tenant's encrypted database.

This approach eliminates an entire class of data exposure risk that exists with cloud transcription services like Rev.ai, Assembly AI, or Deepgram — providers that would receive copies of your executive communications for processing.

What Gets Transcribed

Transcription is applied to all audio-bearing communication: video meetings and voice/SMS platform call recordings.

• Microsoft Teams video meeting recordings
• Zoom Meetings cloud recordings
• Google Meet recordings (via Google Workspace connector)
• RingCentral voice call recordings
• Dialpad call recordings
• Zoom Phone call recordings
• OpenPhone call recordings

Language & Translation

The PanOps transcription engine is natively multilingual — it can transcribe and translate audio in over 90 languages in a single pass. PanOps uses this capability to support international teams: if employees speak in Spanish, Mandarin, French, or any supported language, their audio is automatically transcribed and translated to the CEO's configured preferred language.

Language preference is configured per customer during onboarding and stored in the customer's configuration record. All transcripts delivered to the AI model are in the CEO's preferred language.

Pipeline Overview

The transcription pipeline is event-driven and fully automated. When a recording is downloaded to customer storage, an S3 event triggers the transcription queue. GPU workers pick up jobs, run the PanOps transcription engine, and write the completed transcript directly to the customer's secure database. The entire process requires no manual intervention.

GPU instances scale to zero when no recordings are queued, keeping compute costs near zero during periods of low or no recording volume. Estimated cost is $50–150/month depending on the volume of meetings and calls.