Voice Agent 🚀

Overview 🌟

This project is an AI-driven conversational voice assistant built using LiveKit Agents and LLaMA Index. It leverages Retrieval-Augmented Generation (RAG) with a vector database to provide context-aware responses, optimized for efficient voice interactions. The assistant selectively retrieves document context based on trigger words, ensuring fast responses for simple queries.

Key features:

• Selective RAG Retrieval: Uses trigger words to determine when to retrieve document context.
• Customizable Trigger Words: Easily adjust which queries trigger RAG retrieval.
• Scalable Knowledge Base: Integrates with a vector database for document retrieval.

About the Creator 👨‍💻

Hi, I'm Arjun, an AI Engineer passionate about building intelligent systems that make life easier. I designed and developed this project, integrating cutting-edge technologies like:

• LLaMA Index for efficient document indexing and retrieval.
• LiveKit Agents for seamless voice interaction.
• OpenAI GPT-4o for natural language understanding and generation.
• Zilliz Cloud for scalable vector storage.

Feel free to reach out if you have any questions or want to collaborate on exciting AI projects! 🌐

Prerequisites 📋

• Python 3.8+
• A vector database account (e.g., Zilliz Cloud, Milvus)
• OpenAI API key for LLM (GPT-4o)
• LiveKit Agents for voice interaction
• LLaMA Index for RAG integration

Installation 🛠️

1. Clone the Repository:

   git clone https://github.com/yourusername/voice-agent.git
   cd voice-agent

2. Set Up a Virtual Environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies:

   pip install -r requirements.txt

   Example requirements.txt:

   livekit-agents
   llama-index
   openai
   

4. Set Environment Variables: Create a .env file in the project root and add your API keys:

   OPENAI_API_KEY=your-openai-api-key
   VECTOR_DB_API_KEY=your-vector-db-api-key
   

5. Prepare Instructions File: Ensure a config/prompt.txt file exists with the system prompt for the assistant.

Usage 🎙️

1. Run the Application:

   python voice_server.py console

   This starts the LiveKit Agents worker, initializes the vector database index, and begins listening for voice inputs.

2. Interact with the Assistant:

   • Use a voice client compatible with LiveKit to interact with the assistant.
   • Example queries:
     • "What courses do you offer?"
     • "How much does the data science course cost?"
     • "When is the application deadline?"

3. Monitor Logs: Logs will display RAG decisions (performed or skipped) and retrieved context for debugging.

Project Structure 🗂️

coaching_rag_agent/
├── src/
│   ├── agents/
│   │   └── voice_agent.py          # Core voice agent logic with selective RAG retrieval
│   ├── core/
│   │   ├── config.py               # Configuration and project paths
│   │   └── indexing.py             # Index management and vector store operations
│   ├── vector_store/
│   │   ├── upload_documents.py     # Document upload to vector store
│   │   └── test_retrieval.py       # Test vector store retrieval
│   └── utils/
│       └── cloud_utils.py          # Cloud connection utilities
├── data/
│   └── knowledge_base/             # PDF documents for RAG
├── storage/
│   └── vector_storage/             # Local vector store persistence
├── config/
│   └── prompt.txt                  # System prompt for the assistant
├── tests/                          # Test files
├── voice_server.py                 # Entry point for starting the LiveKit voice server
└── requirements.txt                # Python dependencies

Customization ✨

• Trigger Words: Modify the TRIGGER_WORDS list in src/agents/voice_agent.py to control which queries trigger RAG retrieval. Example:

  TRIGGER_WORDS = [
      "who", "what", "where", "when", "why", "how",
      "tell me", "explain", "describe", "give me",
      "information about", "details on", "facts about",
      "course", "courses", "learning", "study", "training",
      "beginner", "intermediate", "advanced", "syllabus",
      "duration", "fee", "fees", "cost", "price",
      "deadline", "application", "start date", "enrollment",
      "data science", "ai", "ml", "web development", "cybersecurity",
      "python", "certificate", "certification", "support",
  ]

• RAG Parameters: Adjust similarity_top_k in the as_retriever() call to balance retrieval speed and accuracy.

Contributing 🤝

Contributions are welcome! Please follow these steps:

1. Fork the repository.
2. Create a new branch (git checkout -b feature/your-feature).
3. Make your changes and commit (git commit -m "Add your feature").
4. Push to your branch (git push origin feature/your-feature).
5. Open a pull request.

License 📜

This project is licensed under the MIT License. See the LICENSE file for details.