Communication platform that helps non-English speakers communicate with English-speaking recipients through AI-assisted translation and a human relay agent.

• Caller speaks in their language → Speech-to-text → Translated to English → Agent reads to receiver.
• Receiver replies in English → Translated to caller language → Text-to-speech → Caller hears response.

No login; single demo path focused on core relay functionality.

• Frontend: Next.js 14, React, TypeScript
• Backend: Python, FastAPI, WebSocket
• Database: Supabase (call sessions, messages)

HackCan/
  README.md
  frontend/          # Next.js app (agent dashboard, transcript UI)
  Backend/           # FastAPI (speech, translation, sessions, WebSocket)
  shared/            # Shared types and constants

cd Backend
python -m venv venv
# Windows: venv\Scripts\activate
# macOS/Linux: source venv/bin/activate
pip install -r requirements.txt

Create Backend/.env (or project root .env):

SUPABASE_URL=your_supabase_url
SUPABASE_KEY=your_supabase_anon_key

Run the API (from repo root):

cd Backend && python main.py

Or: uvicorn Backend.main:app --reload --app-dir . from repo root (adjust so Backend is on PYTHONPATH). Easiest: from Backend folder run python main.py. Server runs at http://localhost:8000.

cd frontend
npm install
npm run dev

Open http://localhost:3000. Optional env (create frontend/.env.local):

NEXT_PUBLIC_API_URL=http://localhost:8000
NEXT_PUBLIC_WS_URL=ws://localhost:8000/ws

1. Click Start call session on the dashboard.
2. Agent dashboard shows session ID and transcript area.
3. Caller speech (POST /process-audio with base64 audio) → STT → translate to English → transcript updated (and over WebSocket).
4. Agent reads the English text to the receiver.
5. Agent types receiver’s English reply in Receiver reply and clicks Send translated to caller (POST /translate-response).
6. Backend translates to caller language and can return TTS audio for the caller.

For persistence, create tables (e.g. in SQL editor):

• call_sessions: id, status, caller_language, receiver_language, created_at, ended_at
• messages: id, session_id, sender, original_text, translated_text, language_code, created_at

Backend currently keeps sessions in memory; wire backend/db/supabase_client into pipeline if you want DB persistence.

• Shared contract: All shared types live in shared/types.py; everyone imports from there.
• Service interfaces: Voice services expose transcribe_audio, detect_language, translate_text, generate_speech; routes rely on these.
• No cross-folder edits: Frontend does not modify backend services; backend does not modify UI components.

MIT (or your choice).