// case study
Alara Plan
A self-hosted, full-stack app that turns recorded meetings (English or Urdu) into transcripts with speaker diarization, AI-generated Minutes of Meeting, and auto-created Kanban / Scrum tasks.
01 The problem
Teams lose hours turning meetings into action, notes are inconsistent, tasks get forgotten, and recordings just sit unused. Alara Plan set out to:
1. Transcribe meeting audio accurately, including Urdu, with speaker labels
2. Generate structured, reliable Minutes of Meeting automatically
3. Convert decisions into tracked tasks without manual copying
1. Transcribe meeting audio accurately, including Urdu, with speaker labels
2. Generate structured, reliable Minutes of Meeting automatically
3. Convert decisions into tracked tasks without manual copying
02 The solution
An end-to-end pipeline from audio to action:
1. Upload & transcode: recordings stored via MinIO (S3-compatible)
2. Transcription: Deepgram Nova-2 with speaker diarization (English & Urdu)
3. AI Minutes: Claude (via OpenRouter) generates structured Minutes of Meeting
4. Auto-tasks: decisions become cards on Kanban or Scrum boards, drag-and-drop with @dnd-kit
5. Robust backend: Node / Express + Prisma on PostgreSQL, background jobs on Redis + BullMQ, with Slack and email notifications
1. Upload & transcode: recordings stored via MinIO (S3-compatible)
2. Transcription: Deepgram Nova-2 with speaker diarization (English & Urdu)
3. AI Minutes: Claude (via OpenRouter) generates structured Minutes of Meeting
4. Auto-tasks: decisions become cards on Kanban or Scrum boards, drag-and-drop with @dnd-kit
5. Robust backend: Node / Express + Prisma on PostgreSQL, background jobs on Redis + BullMQ, with Slack and email notifications
03 The result
Meetings become searchable records and a ready-to-work backlog in minutes, no manual note-taking, no dropped action items. Built with Next.js 15, React 19, and a fully containerized backend.