Scottish Courts · Public sector and justice
Courtroom Proceedings Transcription
Delivery context
Personally delivered by Raga AI founder Usman Bukhari through Usman Bukhari Ltd before establishing RAGA AI LTD.
- Sector
- Public sector and justice
Multi-party courtroom audio needed to become a durable, searchable written record of proceedings.
Business challenge
Sensitive audio cannot leave controlled infrastructure, and unscripted overlapping speech defeats generic transcription — while long hearings make per-minute API pricing untenable.
What was delivered
- Whisper large-v3 self-hosted on local GPU with faster-whisper (CTranslate2) for batched throughput
- Speaker diarisation with pyannote.audio across unscripted, overlapping multi-party proceedings
- Ingest, segment, transcribe and store pipeline producing retrievable structured transcripts
- On-premise by design: no audio egress, no per-minute cost on long hearings
Technical implementation
Technology stack
- Whisper large-v3
- faster-whisper
- pyannote.audio
- On-prem GPU
Related service: AI implementation services
Facing something similar?
Request a discovery conversation and we will talk through your workflow, the systems involved and whether this approach would work for you.