ASC Chatbot - A customer support chatbot using LLM + Document Retriever (RAG) in Vietnamese

📝 Project Description

The Customer Support Chatbot is an automated customer support system built using Retrieval-Augmented Generation (RAG) architecture combined with Gemini LLM and LlamaIndex. This chatbot is designed to enhance customer support efficiency, reduce operational costs, and improve the accuracy of responses to customer queries.

🎯 Objectives

• Reduce support team size by 30% through automated customer query handling.
• Lower API costs by 40% using a combination of semantic embeddings and keyword filtering.
• Achieve 85% retrieval accuracy and 95% uptime, with response times under 5 second.

🛠 Technologies Used

• Programming Language: Python
• AI Models: Gemini LLM, RAG Architecture, LlamaIndex
• Natural Language Processing (NLP): Sentence-Transformers
• Embedding Model: ASC Embedding (Fine-tuned)
• Sparese Model: Qdrant/bm42-all-minilm-l6-v2-attentions (HuggingFace)
• Rerank Model: cross-encoder/ms-marco-MiniLM-L-6-v2 (HuggingFace)
• Vector Database: Qdrant
• Backend: FastAPI, Docker
• Deployment: RESTful APIs, Docker Container
• Other Tools: Git, Postman

Workflow

Demo

• Video Demo: https://storage.googleapis.com/ascchatbot/source/ASC.mp4
• Live front-end (no server): https://khavanw.github.io/Customer-Support-Chatbot/
• Homepage
• Chatbot

🚀 Key Features

• 🔹 Multi-Source Data Support

  • The chatbot can process data from various formats, including PDF, Word, and Excel.
  • Data is chunked and converted into JSON format for efficient processing.

• 🔹 Dual Embedding System

  • Uses semantic embeddings (fine-tuned model) and sparse embeddings (keyword matching) to enhance search accuracy.
  • Embedding Model: ASC_Embedding is fine-tuned to match project-specific data.

• 🔹 Hybrid Search Mechanism

  • Combines semantic search and keyword search for optimized query results.
  • Utilizes a reranking model to improve result accuracy.

• 🔹 Gemini LLM Integration

  • Leverages Gemini 1.5 with custom prompt templates for generating natural and precise responses.

• 🔹 Easy Deployment

  • The application is containerized using Docker and deployed as RESTful APIs with FastAPI.

📂 Project Structure

Customer-Support-Chatbot/
├── backend/ # Backend source code
│ ├── data/ # Input data (PDF, Word, Excel)
│ ├── models/ # Fine-tuned AI models
│ ├── src/ # Main source code
│ │ ├── data_processing/
│ │ ├── embeddings/ 
│ │ ├── search/ 
│ │ ├── api/ 
│ │ └── utils/ 
│ ├── Dockerfile 
│ └── requirements.txt 
├── frontend/ # Frontend source code
│ ├── public/ 
│ ├── src/ 
│ │ ├── components/ 
│ │ ├── pages/ 
│ │ ├── App.js 
│ │ └── index.js 
│ ├── Dockerfile 
│ └── package.json 
├── docker-compose.yml # Docker Compose configuration for both Frontend & Backend
└── README.md # Project documentation

🛠 Installation & Running the Project

1️⃣ Setup Environment

Ensure you have Python 3.8+, Node.js, and Docker installed.

Clone the repository (if you have access):

git clone https://github.com/Khavanw/Customer-Support-Chatbot.git
cd Customer-Support-Chatbot

2️⃣ Install Backend Dependencies

Navigate to the backend directory and install required dependencies:

cd backend
pip install -r requirements.txt

Download the fine-tuned Embedding Model:

from transformers import AutoModel
model = AutoModel.from_pretrained("vankha/asc_embedding")
model.save_pretrained("models/asc_embedding")

3️⃣ Install Frontend Dependencies

Navigate to the frontend directory and install required packages:

cd ../frontend
npm install

4️⃣ Run the Project with Docker Compose

Run both the Frontend and Backend using Docker Compose:

docker-compose up --build

5️⃣ Access the Application

• Frontend: http://localhost:3000
• Backend API: http://localhost:8000

📞 Contact

If you have any questions or contributions, feel free to reach out:

• Email: nguyenvankha0812@gmail.com
• GitHub: Khavanw

📜 License

This project is private and not publicly distributed.
All source code and model details are owned by the author.