healthcare chatbot

Overview

This project implements a Django-based Patient Chat Application where patients can interact with an AI bot to manage health-related conversations, request changes to appointments, medications, and access important medical information. It uses a knowledge graph and an LLM-agnostic design to enhance functionality and scalability.

Key Features

• Health-Related Conversations: The bot assists with medical condition inquiries, medications, diets, and appointments.
• Appointment Requests: Patients can reschedule appointments, which are updated for the doctor in real time.
• Entity Extraction: Extracts key entities (e.g., medications, symptoms, appointments) and stores them in a Neo4j knowledge graph for future interactions.
• Memory Optimization: Manages long conversations by storing data in PostgreSQL and Neo4j, ensuring real-time processing.
• LLM-Agnostic Design: Built with Langchain for compatibility with various LLMs, such as Google Gemini, GPT-3, or LLaMA.

Details About Machine Learning

• Classifier for Query Filtering: Binary classifier trained on MedQuAD and Natural Questions datasets, along with a topic model from HuggingFace to filter out irrelevant queries.
• Entity Extraction for Knowledge Graph: Key entities like medications and appointment preferences are extracted using Bio_ClinicalBERT and regex matching.

Project Structure

├── datasets
├── patient_chat_app
│   ├── chatbot
│   ├── ml_models
│   ├── static
│   └── templates
└── venv

Setup Instructions

Follow these steps to set up the application:

1. Clone the repository and set up a virtual environment.
2. Install dependencies with pip install -r requirements.txt.
3. Configure PostgreSQL and Neo4j databases and update settings in settings.py and knowledge_graph.py.
4. Run migrations with python manage.py migrate.
5. Start the server using python manage.py runserver.

Future Improvements

• Implement multi-agent systems to separate task-based models more effectively.
• Add conversation summarization for easier documentation and review.
• Optimize memory storage by reducing redundancy between PostgreSQL and Neo4j.

Conclusion

This project demonstrates advanced chatbot functionality with health-related entity extraction, flexible LLM integration, and real-time appointment management. The architecture ensures scalability and adaptability for various use cases.