🧠 T5-Based Medicine Information Extractor

This project fine-tunes a T5 Transformer model to predict the Uses and Side Effects of medicines based on their name or salt composition.

---

📊 Task Description

• Input: Medicine name or salt composition
  (e.g., "Azithromycin (500mg)")

• Output: (Uses: Treatment of Bacterial infections; Side effects: Nausea Abdominal pain Diarrhea)

---

🏗️ Model Architecture

• Model: t5-small
• Library: 🤗 HuggingFace Transformers
• Training Framework: PyTorch
• Data Size: ~11,000 medicine records
• Format: Custom dataset with fields:
• Composition (input)
• Uses (output)
• Side effects (output)

---

📁 Model Files

t5_medicine_model/
├── config.json
├── pytorch_model.bin
├── tokenizer_config.json
├── special_tokens_map.json
├── spiece.model
├── generation_config.json

🔗 Download from Google Drive

📦 How to Use the Model

from transformers import T5Tokenizer, T5ForConditionalGeneration

# Load from local path (after downloading from Google Drive)
model_path = "./t5_medicine_model"

tokenizer = T5Tokenizer.from_pretrained(model_path)
model = T5ForConditionalGeneration.from_pretrained(model_path)
model.eval()

# Inference
input_text = "Azithromycin (500mg)"
inputs = tokenizer(input_text, return_tensors="pt")

output_ids = model.generate(
    **inputs,
    max_length=128,
    decoder_start_token_id=model.config.decoder_start_token_id
)

result = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print("Prediction:", result)

🧠 Future Work

• Deploy as an API using FastAPI or Flask
• Add a web UI using Gradio or Streamlit
• Improve accuracy using larger Transformer models
• Add confidence scores or implement partial matching

🧑‍💻 Author

Furkan Khan
Model trained locally using Python, HuggingFace Transformers, and PyTorch

📜 Certification

This project was built and trained by Khan F as part of a personal initiative to explore Natural Language Processing using:

• ✅ Python
• ✅ HuggingFace Transformers
• ✅ PyTorch

All training and testing were performed locally.
This repository serves as a showcase of model development, deployment planning, and ongoing improvements in NLP.

📌 This project is not officially certified by any institution and is intended for educational and demonstrative purposes.