oss project

meeting-note-taker

Self-hosted Fireflies.ai - upload audio, get transcripts, summaries, and action items.

Turborepo
React 18
TypeScript
Vite
Tailwind CSS
Shadcn UI
TanStack Query
Socket.io
Express
BullMQ
Drizzle ORM
Supabase
OpenAI Whisper
Google Gemini
Redis
Docker

Turborepo monorepo full-stack app that transcribes uploaded meeting audio with OpenAI Whisper (Google Gemini fallback), summarizes content, and extracts action items.
Real-time WebSocket progress streaming via Socket.io, Google OAuth, BullMQ job queue with Redis, Drizzle ORM against Supabase Postgres.
Full-stack TypeScript end-to-end: React 18 + Vite + Tailwind + Shadcn on the front; Express + BullMQ + Drizzle + Socket.io on the back; Supabase auth/storage; Docker for local dev.

Architecture

apps/
  web/           React 18 + Vite + Tailwind + Shadcn UI + TanStack Query
  api/           Express + BullMQ workers + Socket.io + Drizzle ORM
packages/
  shared/        Zod schemas + types
infra/
  supabase/      Auth + Postgres + Storage
  docker/        Compose stack for local dev

Flow:

User uploads audio → API persists to Supabase Storage, enqueues a transcription job in BullMQ.
Worker pulls the job, runs Whisper (with Gemini fallback for long-form), streams progress events over Socket.io.
On completion, summarization + action-item extraction runs against the transcript.
UI lights up in real time — no polling, no page refresh.

Built primarily as a personal Whisper testbed - and because paying per-minute for meeting transcription felt wrong.