MedWise 🧠💊

MedWise is an AI-powered medical literature and drug interaction analyzer.
On MongoDB Track for AI in Action: Google Cloud hackathon built with Google Cloud Vertex AI and MongoDB Atlas.

💡 Project Overview

MedWise is an innovative AI-powered medical research assistant designed to empower both healthcare professionals and individuals. It bridges the gap between complex medical literature and actionable insights by analyzing vast datasets from PubMed research papers and FDA drug databases. Our core innovation lies in leveraging advanced AI to provide real-time drug interaction analysis, suggest cutting-edge treatments based on the latest research, and democratize health knowledge for everyone.

Whether you're a busy clinician seeking the most current evidence, or a patient trying to understand a new medication, MedWise aims to provide accurate, accessible, and evidence-backed information.

🔍 What It Does

• Analyze drug interactions using LLMs
• Semantic search over PubMed-style medical articles
• Provide evidence-based treatment suggestions
• Offer simple explanations for patients

✨ Key Features

For Healthcare Professionals (Clinical Mode):

• Vector Search for Semantic Retrieval: Go beyond keyword matching. Ask nuanced questions, and MedWise finds conceptually similar research papers and drug data.
• AI-Powered Drug Interaction Analysis: Input patient medications and conditions; MedWise intelligently analyzes potential drug-drug and drug-condition interactions.
• Evidence-Based Treatment Suggestions: Get AI-generated summaries of the latest treatment protocols for specific conditions, directly citing original research papers.
• Confidence Scoring: Understand the relevance and reliability of suggested information.

For Patients & Caregivers (MyMedWise - Novice Mode):

• Simple Language Explanations: Translate complex medical jargon into easy-to-understand terms for drugs, diagnoses, and health concepts.
• Common Interaction Clarification: Get simplified explanations of potential common drug interactions relevant to daily life.
• Proactive Health Literacy: Empower users to ask informed questions and better understand their health journey.
• Crucial Disclaimer: Always provides clear disclaimers that information is for educational purposes only and not a substitute for professional medical advice.

🚀 Technology Stack

MedWise is built on a powerful combination of cloud-native and AI technologies:

• Google Cloud AI:
  • Vertex AI Embeddings API for Text: Used for generating high-quality vector embeddings from medical literature and user queries, crucial for semantic search.
  • Vertex AI Generative AI Studio (Gemini API): The core intelligence behind our natural language understanding, drug interaction analysis, treatment suggestion synthesis, and simplified explanations.
• MongoDB Atlas:
  • The flexible and scalable database for storing vast amounts of PubMed articles, FDA drug data, and their associated vector embeddings.
  • MongoDB Atlas Vector Search: Enables efficient and lightning-fast semantic search based on vector similarity.
• More:

• React.js (Frontend)
• MongoDB Atlas + Vector Search
• Google Cloud Vertex AI (Embeddings)
• Gemini (LLM Responses)
• Axios, Lucide Icons
• Streamlit (optional fallback UI)

🛠️ Setup & Local Development

To get MedWise running locally or understand its architecture:

1. Clone the Repository:

   git clone https://github.com/darwinwatts/medwise
   cd medwise

2. Google Cloud Setup:
   • Ensure you have a Google Cloud project with billing enabled.
   • Enable necessary APIs: Vertex AI API, Cloud Natural Language API.
   • Set up authentication (e.g., gcloud auth application-default login).
3. MongoDB Atlas Setup:
   • Create a free-tier MongoDB Atlas cluster.
   • Obtain your connection string.
   • Create [your_database_name] and collections (e.g., pubmed_articles, fda_drugs).
   • Important: Create a Vector Search index on your pubmed_articles collection for the embedding field.
4. Data Ingestion & Embedding:
   • Run the data ingestion script: python scripts/ingest_data.py (This script will download/parse a sample of PubMed/FDA data, generate embeddings using Vertex AI, and load them into MongoDB Atlas).
5. Backend Deployment:
   • cd backend
   • pip install -r requirements.txt (for Python) or npm install (for Node.js)
   • python app.py (or node server.js)
   • For Cloud Run/Functions deployment: Follow Google Cloud's documentation for your chosen language.
6. Frontend/Demo App:
   • cd frontend
   • pip install -r requirements.txt (if using Streamlit)
   • streamlit run app.py
   • Access the app in your browser at http://localhost:8501 (or your deployed URL).

📊 Demo Walkthrough

*(The demo video and explaination for judges transcript)

1. Clinical Mode - Query for Latest Treatments:

   • Navigate to the Clinical Mode interface.
   • Input a complex query: "What are the latest treatments for type 2 diabetes in elderly patients with kidney impairment?"
   • Observe MedWise retrieving relevant papers using semantic search (powered by Vertex AI Embeddings & MongoDB Atlas Vector Search).
   • See the AI-synthesized answer (from Gemini) with clear citations to original PubMed IDs/titles.

2. Clinical Mode - Drug Interaction Analysis:

   • Input a list of drugs: "Analyze interactions between Metformin, Simvastatin, and a new anti-inflammatory drug, Ibuprofen, for a patient with kidney impairment."
   • MedWise uses Gemini to analyze FDA data and research, highlighting potential risks.

3. MyMedWise (Virtual Patient Mode) - Simple Explanation:

   • Switch to the "MyMedWise" or "Virtual Patient Mode" interface.
   • Input a common question: "Explain 'hypertension' in simple terms, like you're talking to a friend."
   • See MedWise (powered by Gemini) rephrase complex medical definitions into understandable language, along with the crucial medical disclaimer.

• Devpost Submission

🤝 Contributing

While this is a solo hackathon project, future contributions would be welcome!

🚀 Future Enhancements

• Real-time Literature Updates: Integrate with PubMed APIs for continuous ingestion of new research.
• User Feedback Loop: Allow users to rate the helpfulness of answers to refine AI models.
• Multi-Modal Inputs: Support image/chart analysis in medical papers.
• Integration with EMR/EHR Systems: (Complex, long-term) Seamlessly connect with electronic health records for personalized patient context.
• Personalized Learning Paths: For healthcare professionals, suggest learning modules based on their queries.

📜 License

This project is licensed under the MIT License.