🏥 MediBot AI — My Medical AI Assistant

---

📌 Table of Contents

• Overview
• Live Demo
• Features
• Model Details
• Tech Stack
• Project Structure
• How It Works
• Getting Started
• Disclaimer
• License

---

🧠 Overview

MediBot AI is a domain-specific medical chatbot built by fine-tuning TinyLlama-1.1B-Chat on a custom medical_qa.csv dataset using QLoRA (Quantized Low-Rank Adaptation). It answers medical questions covering symptoms, diseases, drug information, general health advice, and more — all through a clean Gradio web interface deployed on HuggingFace Spaces.

This project demonstrates a full LLM fine-tuning pipeline: from dataset preparation → QLoRA training → adapter merging → Gradio deployment.

---

🌐 Live Demo

🤗 https://huggingface.co/spaces/Prajwal4107k/My-Medical-AI-Assistant

💬 MediBot AI — Symptom Query

Asking "What are the symptoms of diabetes?" — MediBot responds with a detailed, numbered breakdown of diabetes symptoms including hyperglycemia, numbness, blurred vision, and fatigue.

---

🩺 MediBot AI — Multi-turn Medical Q&A

Multi-turn conversation on HuggingFace Spaces — the bot correctly answers "What is the average human temperature?" and provides detailed kidney disease symptom identification when asked.

---

✨ Features

• 🤖 Fine-tuned LLM — TinyLlama 1.1B Chat, fine-tuned specifically on medical Q&A data
• 💊 Medical Q&A — Answers questions about symptoms, diseases, and conditions
• 🌡️ Symptom Checker — Identifies possible conditions based on described symptoms
• 💉 Drug Information — Provides general information about medications
• 🩺 General Medical Advice — Health tips, normal ranges, body metrics
• 🖥️ Gradio Web UI — Clean, dark-themed chat interface with example prompts
• 🤗 HuggingFace Spaces — Publicly accessible live deployment
• ⚡ QLoRA Training — Efficient fine-tuning with 4-bit quantization using PEFT

---

🧪 Model Details

---

🛠️ Tech Stack

---

📁 Project Structure

My-Medical-AI-Assistant-Code/
│
├── fine_tuned_model/                   # Trained LoRA adapter files
│   ├── adapter_config.json             # LoRA adapter configuration
│   ├── adapter_model.safetensors       # Fine-tuned adapter weights
│   ├── chat_template.jinja             # Custom chat prompt template
│   ├── special_tokens_map.json         # Special tokens mapping
│   ├── tokenizer.json                  # Tokenizer vocabulary
│   ├── tokenizer.model                 # SentencePiece tokenizer model
│   ├── tokenizer_config.json           # Tokenizer configuration
│   └── README.md                       # Model card
│
├── .gradio/
│   └── certificate.pem                 # Gradio SSL certificate
│
├── app.py                              # Gradio app — chat interface
├── fine_tune_qlora.py                  # QLoRA fine-tuning script
├── medical_qa.csv                      # Medical Q&A training dataset
├── dataset.json                        # Processed dataset in JSON format
├── test_imports.py                     # Dependency verification script
├── requirements.txt                    # Python dependencies
└── README.md

---

⚙️ How It Works

medical_qa.csv  →  dataset.json
      │
      ▼
┌──────────────────────────────┐
│   fine_tune_qlora.py         │
│                              │
│  • Load TinyLlama-1.1B-Chat  │
│  • 4-bit quantization        │
│  • Attach LoRA adapters      │
│  • Train on medical Q&A      │
│  • Save adapter weights      │
└──────────────┬───────────────┘
               │
               ▼
    fine_tuned_model/
    adapter_model.safetensors
               │
               ▼
┌──────────────────────────────┐
│         app.py               │
│                              │
│  • Load base model           │
│  • Merge LoRA adapter        │
│  • Apply chat template       │
│  • Launch Gradio UI          │
└──────────────┬───────────────┘
               │
               ▼
    ┌─────────────────────┐
    │   MediBot AI        │  ←  User asks medical question
    │   Gradio Chat UI    │  →  LLM generates answer
    └─────────────────────┘
               │
               ▼
    HuggingFace Spaces 🤗
    (Public live deployment)

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• GPU recommended (CUDA) for inference; CPU works but is slower
• HuggingFace account (for model access)

---

1. Clone the Repository

git clone https://github.com/prajwal5065/My-Medical-AI-Assistant-Code.git
cd My-Medical-AI-Assistant-Code

---

2. Install Dependencies

pip install -r requirements.txt

Core packages include:

transformers
peft
bitsandbytes
gradio
torch
accelerate
datasets

---

3. Verify Imports

python test_imports.py

---

4. (Optional) Re-run Fine-tuning

python fine_tune_qlora.py

⚠️ Requires a CUDA GPU. The script loads medical_qa.csv, applies 4-bit quantization, trains LoRA adapters, and saves them to fine_tuned_model/.

---

5. Launch the Gradio App

python app.py

The MediBot AI interface will open at http://127.0.0.1:7860

---

6. Or Use the Live Demo

No setup needed — just visit:

🤗 https://huggingface.co/spaces/Prajwal4107k/My-Medical-AI-Assistant

---

⚠️ Disclaimer

MediBot AI is for informational and educational purposes only. It is NOT a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare professional for medical decisions. Never ignore professional medical advice because of something this AI has said.

---

📄 License

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

---