MedlinePlus Chatbot – AI-Powered Clinical Knowledge Assistant

AI-driven clinical chatbot that references trusted medical content from MedlinePlus through a custom Retrieval-Augmented Generation pipeline backed by a high-performance vector database to deliver concise, evidence-backed medical answers in real time.

Overview

• React + TailwindCSS - Modern, responsive web interface built with Vite for speed
• FastAPI Backend - API layer
• Custom RAG Pipeline – Parses and indexes MedlinePlus XML into a local FAISS vector database
• FAISS + SentenceTransformers – L2 Semantic search with cached embeddings for sub-second retrieval
• Evidence-Backed Responses - Every answer cites ranked MedlinePlus sources with relevance scores
• LLM Integration – Works seamlessly with OpenAI, DeepSeek, Groq, or any compatible API
• Local Knowledge Base - Operates independently without external databases after setup

Data Source: MedlinePlus

This chatbot leverages MedlinePlus — a health information service of the U.S. National Library of Medicine (NLM), part of the National Institutes of Health (NIH). MedlinePlus provides high-quality, reliable, and up-to-date information on diseases, conditions, medications, and wellness topics. All content is evidence-based, written in plain language, and reviewed by medical experts.

For more details, visit medlineplus.gov.

Project Structure

/frontend        # React-based UI (Vite + TailwindCSS)
/server          # FastAPI backend (RAG + LLM orchestration)
/server/mplus_*.xml  # MedlinePlus XML medical dataset

Backend Setup

1. Navigate to backend:

   cd server

2. (Optional) Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set your API key securely: Create a .env file in /server:

   OPENAI_API_KEY=your-api-key

5. Add an updated MedlinePlus XML dataset (optional): Place .xml file (e.g. mplus_topics_2025-07-19.xml) in /server.

6. Start the backend:

   uvicorn main:app --reload

First run will:

• Parse MedlinePlus XML topics
• Generate embeddings
• Build a FAISS index
• Cache as faiss_index.bin and documents.json

Backend runs at: http://127.0.0.1:8000

Frontend Setup

1. Navigate to frontend:

   cd frontend

2. Install dependencies:

   npm install

3. Start development server:

   npm run dev

Frontend runs at: http://localhost:5173

Usage

1. Open the app in your browser.
2. Ask a clinical question, e.g.: "What are the symptoms of pneumonia?"
3. CuraLinkAI will:
   • Retrieve relevant content from the FAISS index
   • Pass it to the LLM for synthesis
   • Return a clear, concise answer with cited sources

(If the image appears blurry in this preview, please click on it to view in full resolution)

Future Enhancements

• Customized RAG implementation from scratch
• Add more sources
• Offline functionality
• Advanced filtering by specialty or evidence level
• Dockerized setup for streamlined deployment and reproducibility

Why RAG Matters

Retrieval-Augmented Generation (RAG) combines the strengths of information retrieval and large language models to deliver accurate, trustworthy, and explainable answers.

Instead of relying solely on the LLM's internal knowledge — which can be outdated or hallucinated — RAG fetches up-to-date, domain-specific information from a trusted source (in this case, MedlinePlus) before generating a response.

This approach ensures:

• Accuracy – Answers are grounded in verified medical resources
• Explainability – Responses include citations to original sources
• Adaptability – Knowledge base can be updated without retraining the model
• Reduced Hallucination – Minimizes fabricated or misleading medical advice

Disclaimer

This project is for experimental and research purposes only. It is not a substitute for professional medical advice.