Real-Time AI Sales Agent (Voice) 🎙️

A production-grade voice AI agent capable of holding natural sales conversations with <500ms latency.

Tech Stack:

• Inference: Cerebras (LLaMA 3.3 70B) - Instant token generation.
• Transport: LiveKit - Real-time WebRTC infrastructure.
• STT: Deepgram - Nova-2 Speech-to-Text.
• TTS: Cartesia - Sonic Low-latency Text-to-Speech.
• VAD: Silero Voice Activity Detection.

Architecture

The agent runs a continuous loop:

1. Listen: Ingest audio stream via LiveKit/Deepgram.
2. Think: Process query with LLaMA 3.3 on Cerebras Wafer-Scale Engine.
3. Speak: Stream text to Cartesia for instant audio synthesis.

Usage

This project is designed to run in Google Colab or a local Python environment.

1. Install dependencies:

   pip install livekit-agents livekit-plugins-openai livekit-plugins-deepgram livekit-plugins-silero

2. Set API Keys:

   • CEREBRAS_API_KEY
   • LIVEKIT_API_KEY
   • DEEPGRAM_API_KEY
   • CARTESIA_API_KEY

3. Run the notebook virtual_sales_agent.ipynb.

---

I used Antigravity to automate this ReadMe!