📞 Call Center Chatbot

An AI-powered Call Center Chatbot using completely open-source models that can run on CPU and GPU. This chatbot handles customer inquiries, retrieves company policies, and accesses customer data efficiently.

---

✨ Features

✅ Uses fully open-source models like Mistral-7B-Instruct (GGUF)
✅ Runs locally on CPU or GPU—no external APIs
✅ Supports customer data lookup via SQL database
✅ Fetches company policies dynamically
✅ Built with Flask for easy web deployment

---

🚀 Installation & Setup

1️⃣ Clone the Repository

git clone https://github.com/alishbalaeeq/CallCenterChatbot.git
cd CallCenterChatbot

2️⃣ Create a Virtual Environment

conda create --name chatbot python=3.10 -y  # For Conda users
conda activate chatbot
# OR
python -m venv chatbot  # For venv users
source chatbot/bin/activate  # macOS/Linux
chatbot\Scripts\activate  # Windows

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Set the Model Path as an Environment Variable

Before running, define the model path (GGUF format):

🔹 Linux/macOS

export MODEL_PATH="/absolute/path/to/mistral-7b-instruct-v0.1.Q4_K_M.gguf"

🔹 Windows (PowerShell)

$env:MODEL_PATH="C:\absolute\path\to\mistral-7b-instruct-v0.1.Q4_K_M.gguf"

---

🛠️ Running the Chatbot

python app.py

The chatbot will be available at http://127.0.0.1:5000/ in your browser.

---

🧠 Model & LLM Setup

This project exclusively uses open-source AI models to ensure privacy, transparency, and full local control.
It leverages:

• Mistral-7B-Instruct (GGUF format) from TheBloke
• Runs via LlamaCpp, enabling execution on CPU or GPU
• No proprietary APIs—everything runs locally

from langchain_community.llms import LlamaCpp
import os

# Load model from environment variable
model_path = os.getenv("MODEL_PATH")

llm = LlamaCpp(
    model_path=model_path,
    temperature=0.1,
    max_new_tokens=256,
    context_window=3900,
    n_ctx=2048,
    n_gpu_layers=50,  # Set to 0 for CPU-only
    verbose=True
)

⚡ Want to Run on CPU?

If you don't have a GPU, simply set:

n_gpu_layers = 0  # Forces LlamaCpp to run entirely on CPU

This ensures that it runs smoothly even on CPU-only machines.

---

📂 Project Structure

CallCenterChatbot/
│── files/                     # Stores database & company policies
│── static/                    # CSS and frontend assets
│── templates/                 # HTML templates
│── utils/                     # Modular utility scripts
│   ├── config.py              # Logging & global settings
│   ├── llm_setup.py           # LLM initialization (LlamaCpp)
│   ├── tools.py               # Tools for database & policy queries
│── app.py                     # Main Flask application
│── requirements.txt           # Dependencies list
│── README.md                  # You’re reading it now!

---

🛠️ Tools & Technologies

• Flask → Web Framework
• LangChain → AI Agent Orchestration
• LlamaCpp → Efficient local model inference
• SQLite → Customer data storage
• Hugging Face Transformers → Model loading

---

💡 Customization

Want to use a different open-source model? Modify the model path:

export MODEL_PATH="/path/to/another-model.gguf"

Then restart the app!

---

🔒 Privacy & Security

✅ No external API calls—All processing happens locally
✅ No data sharing—Ensures customer privacy
✅ Fully Open-Source—No vendor lock-in

---

📬 Contact & Contributions

We welcome contributions! Feel free to fork, submit issues, or suggest improvements. 🚀

---

🔥 Call Center Chatbot → AI-driven customer support using 100% open-source models! 🔥