RAG4HealthQA

• 

  Result 1

• 

  Result 2

• 

  Result 3

• 

  Result 4

🩺 About the Project – RAG4HealthQA

🎯 Inspiration

Healthcare information is everywhere, but it’s often dense, inconsistent, and difficult to understand. Many patients, especially those from non-medical backgrounds, struggle to find clear, trustworthy answers to basic health questions. Inspired by this challenge, we created RAG4HealthQA — an AI-powered assistant that delivers accessible and explainable healthcare knowledge, grounded in content from WHO, NIH, CDC, and other verified public sources.

Our mission: Empower patients and caregivers through AI that informs, not confuses.

💡 Example Questions from Our Knowledge Base

RAG4HealthQA can answer practical, everyday health questions such as:

• What are the symptoms of hypertension?
• How much water should adults drink daily?
• Is a vegetarian diet safe for athletes?
• How does Cognitive Behavioral Therapy (CBT) help with PTSD?
• What are the early signs of diabetes?
• What foods are high in iron?
• What is the DASH diet and who is it for?
• How do I manage anxiety without medication?

These examples reflect a mix of clinical, nutritional, and behavioral health topics, supporting public health literacy and preventive care education.

🛠️ How I Built It

• Frontend: Built using Streamlit to ensure a clean, intuitive UX accessible on both desktop and mobile.
• Backend Architecture: RAG pipeline built using LangChain, FAISS, and Cohere.
• Embeddings + Vector Store: Texts embedded into vector space for fast similarity search via FAISS.
• Data Sources: Public health documents in .md and .txt format sourced from:
  • WHO
  • NIH/NIMH
  • Nutrition.gov
• Modular Pipeline: rag_pipeline.py handles retrieval, LLM completion, and response formatting.

📚 What I Learned

• How to implement domain-specific RAG systems that balance relevance with safety.
• The role of UX simplicity in making technical health data digestible.
• How to curate and chunk real medical content effectively for AI use.
• Adapting non-OpenAI LLMs like Cohere for regulated fields.

⚠️ Challenges I Faced

• Ensuring medical accuracy without overstepping into diagnostic claims.
• Avoiding hallucinations through better prompt engineering and strict document grounding.
• Handling file-based KB refreshes without breaking embeddings or app flow.
• Designing for non-technical users, including older adults and caregivers.

🌍 Why It Matters

RAG4HealthQA bridges the gap between AI innovation and public service. By combining transparency, usability, and trusted knowledge, it supports:

• Digital health equity
• AI safety and explainability
• Public health awareness

This project is part of a broader effort to demonstrate how AI can serve the national interest, aligning with goals under the EB2-NIW category for advanced technologists contributing to U.S. public welfare.

Built With

• .env
• cohere
• faiss
• github
• langchain
• pandas
• python
• streamlit
• streamlitcloud