AI-Powered Healthcare Chatbot

An Intelligent Assistant for Symptom Analysis and Disease Prediction

Overview

This project was developed as part of the GMG (Garje Marathi Global) – MAAI Hackathon, where our team secured 1st place! The AI-powered healthcare chatbot is designed to analyze symptoms provided by users and offer insights into potential diseases, with a focus on early detection of lung cancer in this prototype.

Key Features

• Symptom Analysis: Users can input symptoms, and the chatbot responds with probabilities (high/low) of having a disease.
• Advanced AI Integration: Utilizes Nugen API for generating embeddings and implementing cosine similarity for document retrieval.
• Improved Healthcare Accessibility: Cost-effective solution for patients and healthcare providers to improve early diagnosis, engagement, and satisfaction.
• Streamlit UI: Interactive user interface for seamless interaction.

---

Project Workflow

1. Load Data: Import lung cancer survey data from Kaggle.
2. Preprocess Data:
   • Encode categorical variables.
   • Fill missing data.
3. Train Classifier: Use a Decision Tree Classifier to predict lung cancer.
4. Generate Text Data: Convert data rows into natural language descriptions.
5. Embed Document Texts: Utilize Nugen API for embedding creation.
6. Store Embeddings: Save embeddings in a pickle file for efficient querying.
7. User Query Input: Allow users to input queries about symptoms or diseases.
8. Query Embedding: Convert user queries into embeddings using Nugen API.
9. Cosine Similarity Computation: Compare query embeddings with pre-stored embeddings for relevance ranking.
10. Retrieve and Rank: Identify and rank top-k relevant documents based on similarity scores.
11. Chatbot Response: Provide users with insights on their probability of having lung cancer (High/Low).

---

Technology Stack

• Programming Language: Python
• APIs and Libraries: Nugen API, Pandas, Scikit-learn, Streamlit
• Machine Learning Models: Decision Tree Classifier, SVM, Random Forest, KNN, BERT (for advanced use cases)
• Deployment: Oracle Cloud

---

Getting Started

Prerequisites

1. Install Python 3.8 or above.
2. Install required libraries:

   pip install -r requirements.txt

Running the Application

Frontend

To start the Streamlit user interface, run:

streamlit run streamlit_app.py  

Backend

To launch the backend API using Uvicorn, run:

uvicorn main:app --reload  

Ensure both the frontend and backend are running simultaneously for the application to function properly.

---

Team

• Anupama Aphale
• Chirag Dhamange
• Shahil Dhotre
• Shubham Narkhede
• Rahul Phadtare
• Savani Shrotri

---

Acknowledgments

• Garje Marathi Global and its visionary founder Anand Ganu, along with his dedicated team, for organizing this remarkable event.
• Stanford University for hosting the hackathon.
• Pushkar Nandkar (SambaNova Systems) and Saurabh Netravalkar (Oracle) for their invaluable contributions as sponsors and mentors.
• Kaustubh Supekar (Stanford University) and Niraj Kumar Singh (Nugen) for their exceptional guidance.

---