Customer Support Chatbot (Fine-Tuned DialoGPT)

A fine-tuned conversational AI chatbot designed to handle customer support queries using Microsoft's DialoGPT model. This project adapts a general-purpose language model to a customer service context using a real-world dataset of ~27,000 instruction-response pairs.

🔗 Kaggle Model: Dialogpt-Fine-TunedCustomer-Service

---

Project Highlights

• Fine-tuned on real-world customer support data.
• Built on top of microsoft/DialoGPT-medium.
• Trained using the Bitext Customer Support Dataset.
• Quantized to FP8 precision for easy local deployment and faster inference time.
• Deployed with an easy-to-use Streamlit interface.
• Fully local and offline-ready inference pipeline.
• Modular and readable code structure.

---

Dataset

• Name: Bitext Sample Customer Support Dataset
• Size: 27,000 instruction-response pairs
• Format: CSV with instruction and response columns
• Source: Kaggle Dataset

---

Model Architecture

• Base Model: DialoGPT-medium
• Tokenizer: AutoTokenizer with EOS token as padding
• Training Framework: Hugging Face Transformers
• Loss: Causal LM Loss using language modeling head
• Batch Size: 2
• Epochs: 3

---

Setup Instructions

1. Clone the Repository

git clone https://github.com/<your-username>/customer-support-chatbot.git
cd customer-support-chatbot

2. Install Dependencies

pip install -r requirements.txt

3. Download and place the model files

Download the model files from the kaggle models page mentioned above, and rename it as .model/ and place it in the same directory as the chatbot

4. Run the chatbot locally

streamlit run app.py

Example Usage

User: "Hi,how do I reset my password?" Bot: "Sure! ou can reset your password by clicking on 'Forget Password' at the login screen. Follow the steps sent to your registered email."

Future Work

• Soon to be implemented Retreival Augmented Generation (RAG) for easy customization
• Add support for multi-turn conversations
• Deploy using Gradio or Hugging face spaces
• Integrate feedback loop for real-time improvement
• Extend support for multilingual queries.