feat(transcription): Add Faster Whisper transcription worker foundation and job queue

[dhruvi-16-me]

Summary

This PR introduces the foundation for replacing the existing Vexa-based transcription ingestion pipeline with an internally managed transcription workflow using Faster Whisper.

The implementation adds a dedicated transcription job queue, meeting recording storage infrastructure, and a Python worker capable of processing queued recordings and writing transcripts back to the existing meetings pipeline.

The goal of this phase is to validate that meeting recordings can be transcribed without Vexa while preserving compatibility with the current downstream processing flow (final_transcription, summarization, embeddings, and RAG).

Screen Recording

Google Drive Link

Background

The current meeting intelligence pipeline relies on Vexa to generate transcripts:

Meeting
  ↓
Vexa Bot
  ↓
meetings.transcription
  ↓
final_transcription
  ↓
meeting_summary_json
  ↓
summary_embedding
  ↓
RAG

This dependency introduces external service requirements for transcription.

To enable an internally managed transcription stack, I am introducing a new transcription ingestion layer based on Faster Whisper while keeping all downstream consumers unchanged.

Changes

Database

Added transcription_jobs table

Introduced a dedicated queue table for transcription processing.

Responsibilities:

• Store transcription requests
• Track processing state
• Enable future retry mechanisms
• Decouple recording ingestion from transcription execution

Job lifecycle:

pending
  ↓
processing
  ↓
completed / failed

Added meeting recordings storage support

Created storage infrastructure for uploaded recordings.

Bucket:

meeting-recordings

Path structure:

{team_id}/{meeting_id}/{filename}

This structure aligns with existing team-based access controls and storage policies.

Faster Whisper Worker

Added a new worker:

workers/faster-whisper-worker/

Key files:

worker.py
config.py
requirements.txt

Worker Flow

1. Poll oldest pending transcription job.
2. Claim the job.
3. Download recording from Supabase Storage.
4. Transcribe audio using Faster Whisper.
5. Write transcript to meetings.transcription.
6. Mark job as completed or failed.

Generated transcript format:

{
  "segments": [
    {
      "speaker": "Speaker",
      "text": "..."
    }
  ]
}

This format was intentionally chosen to remain compatible with existing database triggers and processing functions.

Validation

The implementation was validated end-to-end using a real uploaded audio file.

Validation Steps

1. Created test meeting record.
2. Uploaded audio file (test.m4a) to meeting-recordings.
3. Inserted a row into transcription_jobs.
4. Executed:

python worker.py --once

5. Verified:

   • Audio downloaded successfully.
   • Language detection executed successfully.
   • Transcript generated by Faster Whisper.
   • meetings.transcription populated.
   • Job status updated to completed.
   • Existing trigger generated final_transcription.

Verified Pipeline

Meeting Recording
      ↓
Supabase Storage
      ↓
transcription_jobs
      ↓
Faster Whisper Worker
      ↓
meetings.transcription
      ↓
populate_final_transcription
      ↓
final_transcription

No modifications were required to existing transcript consumers.

Future Work

This PR establishes the transcription foundation only.

Upcoming work includes:

• Integrating recording uploads from the Flutter application.
• Automating transcription job creation.
• Validating summarization and embedding generation against Faster Whisper output.
• Replacing remaining Vexa dependencies.

✅ Checklist

• I have read the contributing guidelines.
• I have added tests that prove my fix is effective or that my feature works.
• I have added necessary documentation (if applicable).
• Any dependent changes have been merged and published in downstream modules.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

🗂️ Base branches to auto review (2)

• main
• develop

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: fc91ae37-e468-449f-9e08-956f7d519bb0

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	python-dotenv@​1.2.2
	faster-whisper@​1.2.1
	supabase@​2.31.0

View full report