Production-grade voice AI agent console built for Freya (YC S25). Real-time voice conversation system with LiveKit streaming.

• Voice-to-Voice AI: Groq Whisper STT → llama-3.1-8b → Cartesia TTS
• Real-time Streaming: LiveKit-powered bidirectional audio
• Prompt Management: CRUD operations with versioning
• Session Analytics: Metrics and conversation history
• Production Ready: Dockerized deployment

• Next.js 15 (App Router)
• TypeScript
• Tailwind CSS
• LiveKit Client SDK

• Python 3.11
• LiveKit Agents
• Groq (LLM + STT)
• Cartesia (TTS)
• Silero VAD

• Docker Compose
• Multi-stage builds
• Health checks

• Docker & Docker Compose
• LiveKit server URL + credentials
• Groq API key
• Cartesia API key

1. Clone and configure

git clone <your-repo>
cd <project-folder>
cp .env.example .env

2. Add your API keys to .env

LIVEKIT_URL=wss://your-server.livekit.cloud
LIVEKIT_API_KEY=your_key
LIVEKIT_API_SECRET=your_secret
GROQ_API_KEY=your_groq_key
CARTESIA_API_KEY=your_cartesia_key

3. Build and run

docker compose up --build

4. Access the app

http://localhost:3000

agent/
├── main.py              # Entry point, LiveKit worker
├── requirements.txt     # Python dependencies
└── agent/
    ├── config.py        # Environment config (LLM model, etc)
    ├── voice_agent.py   # Main agent class, LiveKit setup
    └── conversation.py  # Conversation handler (STT → LLM → TTS)

web/
├── app/
│   ├── page.tsx         # Landing/login page
│   ├── api/
│   │   ├── auth/        # Login, logout, check endpoints
│   │   ├── livekit/token/  # LiveKit token generation
│   │   └── prompts/     # Prompt CRUD API
│   └── console/
│       ├── page.tsx     # Main dashboard
│       ├── Components/
│       │   ├── ChatPanel.tsx      # TEXT CHAT (needs work)
│       │   ├── MetricsPanel.tsx   # Metrics display
│       │   ├── PromptModal.tsx    # Create/edit prompts
│       │   └── PromptSidebar.tsx  # Prompt list
│       └── hooks/
│           ├── useLiveKit.ts      # LiveKit connection logic
│           └── usePrompts.ts      # Prompt management
└── lib/
    ├── store.ts         # State management
    └── utils.ts         # Utilities

User Browser
    ↓
Next.js Frontend (Port 3000)
    ↓
LiveKit Cloud
    ↓
Python Agent
    ↓
Groq API (LLM/STT) + Cartesia (TTS)

• STT: Groq Whisper for speech-to-text
• LLM: Groq llama-3.1-8b-instant for responses
• TTS: Cartesia Sonic English voice
• VAD: Silero for voice activity detection

• Create/Read/Update/Delete prompts
• Version history tracking
• In-memory storage (PostgreSQL ready)

• Real-time session tracking
• Conversation history
• Performance metrics
• Analytics dashboard

Backend:

cd agent
pip install -r requirements.txt
python main.py dev

Frontend:

cd web
npm install
npm run dev

# Start services
docker compose up --build

# Run in background
docker compose up --build -d

# View logs
docker compose logs -f

# Stop services
docker compose down

# Fresh start (remove volumes)
docker compose down -v

Focused on core value proposition (voice AI) over text chat to maximize impact in limited timeframe.

Rapid prototyping. Production would use PostgreSQL with Prisma ORM.

Industry standard for real-time voice/video. Powers ChatGPT Advanced Voice Mode.

Smaller images, faster builds, better security with non-root users.

Future enhancements for production deployment:

• PostgreSQL for persistence
• Redis for session caching
• Rate limiting and authentication
• Comprehensive test suite
• Session recording and playback
• Multi-tenant support
• Monitoring and logging (OpenTelemetry)
• Load balancing and scaling
• CI/CD pipeline

Built as Forward Deploying Engineer Assessment for Freya (YC S25)

• Timeframe: 3 days (10/10/25 - 13/10/25)
• Scope: 80% complete (voice working, text chat deprioritized)
• Focus: Production-grade voice AI implementation
• Prioritization: Voice-to-voice pipeline over text chat features

• Real-time bidirectional audio streaming
• Low-latency voice conversation
• Prompt versioning system
• Session analytics and metrics
• Dockerized for portability
• Multi-stage builds for optimization
• Non-root container users for security

• LiveKit: https://livekit.io
• Groq: https://groq.com
• Cartesia: https://cartesia.ai
• Freya (YC S25): https://www.ycombinator.com/companies/freya

This project is licensed under a Portfolio Display License.

TL;DR:

• ✅ You can view and study this code
• ✅ You can reference it in discussions about employment
• ❌ You cannot use this commercially
• ❌ You cannot redistribute or sell this

See LICENSE file for details.

For commercial licensing inquiries: [email protected]

Built by Sanjay Kumar | GitHub | LinkedIn | Portfolio