While the source code isn't public for now, you'll find a detailed breakdown of the architecture, decisions, and implementation approach below.
What
A real-time speech translation system built for IEEE FBU events. A speaker talks into a microphone; attendees connect via QR code and see the translated text live on their phones. The system has two views — speaker and listener — and a Gemini-powered summary tab that condenses the last few sentences.
Why
Built for an IEEE student branch event where speakers and attendees spoke different languages. The goal was a zero-install solution: attendees just scan a QR code.
How it works
...
Features
- Speaker view for audio input and transcript management
- Listener view with live translated text and a Gemini summary tab
- QR code entry — no install required for attendees
- Persistent event storage via Supabase (works on Vercel's ephemeral filesystem)
- n8n workflow for automation/orchestration
- Admin panel for event management
Stack
| Layer | Choice |
|---|---|
| Framework | Next.js + TypeScript |
| Translation / Summary | Gemini API (@google/genai) |
| Database | Supabase (PostgreSQL) |
| Real-time | WebSocket |
| Automation | n8n |
| Styling | Tailwind CSS |
What I learned
...