An open-source, full-stack podcasting platform that records multi-participant video conversations in real-time via WebRTC and automatically produces a fully rendered, editor-ready MP4 using server-side FFmpeg processing. No manual editing required.
Magic Edit: One-Click Post-Production
After a podcast ends, the host clicks Magic Edit. The backend pipeline automatically edits and renders the entire podcast with custom templates and multi-cam layouts, eliminating the need for expensive video editors.
WebRTC Mesh with Resilient Signaling
Real-time, low-latency video between up to 3 participants using WebRTC mesh topology. Custom signaling via Socket.IO handles SDP relay, ICE candidates, health checks, and automatic reconnection.
Production-Grade Auth with 2FA
Google OAuth 2.0 via Passport.js, JWT in HTTP-only cookies (XSS-resistant), CSRF protection, and optional TOTP two-factor auth with QR code enrollment.
Dual Cloud Storage with Progressive Upload
3-second WebM chunks stream to Cloudinary (CDN delivery) and AWS S3 (raw storage). Progressive uploads limit crash data-loss to just 3 seconds.
Live Streaming to YouTube via RTMP
Dockerized Nginx RTMP module relays sessions to YouTube Live or Twitch in real-time, decoupled from recording so stream failures don't affect captures.
Architecture
React App
MediaRecorder · WebRTC · Socket.IO
Socket.IO Server
Signaling · Chat · Sessions
REST API
Auth · Sessions · Recordings
Recording Module
Upload → FFmpeg → Render
MongoDB Atlas
Sessions · Users · Metadata
Cloudinary + S3
Chunks · Final MP4
External Services
OAuth · Gemini AI · SMTP · RTMP
Data Flow
Dashboard: Quick actions hub
Studios Canvas: Session management
Magic Editor: AI-powered post-production
Recording Studio: Live room
Session Details: Metadata & participants
Live Streaming: RTMP relay
Designed the full-stack architecture from scratch: client-server separation, real-time signaling layer, media processing pipeline, and cloud storage strategy
Built the complete WebRTC peer-to-peer signaling system with reconnection handling and connection health checks
Engineered the recording → chunk upload → FFmpeg merge → final render pipeline with Cloudinary/S3 dual-storage
Implemented Google OAuth 2.0 with JWT sessions, HTTP-only cookies, and optional TOTP 2FA
Set up Dockerized Nginx RTMP for live streaming relay to YouTube