AI-Driven Customer Support Agent (Voice & Chat)

Contents

1. Overview
2. Architecture
3. Project Structure
4. Tech Stack
5. Quickstart
6. Ingesting Documents
7. Voice Call Setup
8. Features
9. Deployment

Overview

This project demonstrates a complete production-ready implementation of an AI-powered support assistant. It features multimodal interaction (voice/chat), knowledge retrieval via vector search, and integration with Vapi for real-time voice calling. The system is built with FastAPI, LangChain, Weaviate, and Streamlit, containerized with Docker, and deployable on AWS. These agents handled customer queries with human-like precision across diverse channels, reduced support ticket load, improved CSAT scores, and integrated seamlessly with CRMs and business platforms.

⚙️ Architecture

📁 Project Structure

ai-support-agent/
│
├── app/
│   ├── agents/
│   │   └── support_agent.py
│   ├── chains/
│   │   └── retrieval_chain.py
│   ├── data/
│   │   └── loader.py
│   ├── db/
│   │   └── weaviate_client.py
│   ├── endpoints/
│   │   └── routes.py
│   ├── services/
│   │   └── llm_interface.py
│   ├── vapi/
│   │   └── vapi_integration.py
│   ├── streamlit_app/
│   │   └── ui.py
│   └── main.py
│
├── config/
│   └── settings.py
│
├── tests/
│   └── test_routes.py
│
├── Dockerfile
├── requirements.txt
├── .env
├── README.md

💻 Tech Stack

• Lang: Python, TypeScript, Bash
• Backend/API: FastAPI
• LLMs & Prompt Orchestration: OpenAI (GPT-4), Claude, LangChain,
• Knowledge Retrieval & RAG: Weaviate (primary vector store), LlamaIndex (for document parsing and indexing)
• Voice & Audio Interfaces: Vapi.ai
• Workflow Automation & Orchestration: n8n (for business logic workflows), LangChain Agents (for tool-based tasks)
• Observability & Tracing: LangSmith, LLMGuard (safety filters & evaluation)
• Frontend: Streamlit
• Deployment & Infrastructure: Docker, Terraform, GitHub Actions, AWS
• Security & Compliance: HashiCorp Vault (secrets), OPA (policy), OAuth2, PII masking
• 3rd-Party Integrations: SendGrid, Slack, Google Calendar (for CRM, alerts, reminders)

🚀 Quickstart

1. Clone the Repo

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

2. Set Environment Variables

Create .env file and include:

OPENAI_API_KEY=your-key
WEAVIATE_URL=http://localhost:8080
VAPI_API_KEY=your-vapi-key

3. Install & Run (Locally)

docker-compose up --build

4. Access the App

API: http://localhost:8000
Chat UI: http://localhost:8501

📥 Ingesting Documents

Drop PDFs or .txt files into a folder and run:

python app/ingestion/document_ingestor.py

📞 Voice Call Setup

Make sure your Vapi account is configured correctly. Voice calls can be handled in:

app/vapi/voice_router.py

🧠 Features

• Context-aware, memory-capable chat
• RAG (Retrieval-Augmented Generation) for domain-specific queries
• Voice interaction using Vapi
• Deployable locally or to the cloud
• Easily extendable with more agents, tools, or endpoints

📦 Deployment

• Use the included Docker setup to deploy the API, Weaviate, and Streamlit UI.
• For AWS: Containerize with ECR or ECS
• Attach persistent volume to Weaviate if needed
• Secure API with IAM/SSL/Gateway