CuraLink AI

Surgical precision in clinical insight.

An AI-powered Medical Research Assistant built for complex clinical synthesis.

🎯 Overview

CuraLink is a full-stack, AI-powered Medical Research Assistant designed to move beyond generic chatbots. It is a true health research companion that understands explicit patient context, retrieves high-quality medical literature from primary databases, reasons over immense amounts of data, and delivers structured, highly personalized, and source-backed clinical insights.

Built on the MERN stack and powered by Open-Source LLMs (via Groq/Ollama), CuraLink ensures that medical reasoning remains transparent, verifiable, and blistering fast.

---

🚀 Key Features

• 🧑‍⚕️ Patient-Centric Context: CuraLink establishes explicit Patient Profiles (Age, Condition, Location, Clinical Notes). Every research query is dynamically personalized based on the active patient profile.
• 🧠 Intelligent Query Expansion: Natural language queries (e.g., "Parkinson's DBS") are intercepted by the LLM and structurally expanded into hyper-optimized Boolean strings to maximize retrieval quality from external databases.
• ⚡ Parallel "Scatter-Gather" Retrieval: Interrogates PubMed, OpenAlex, and ClinicalTrials.gov concurrently in real-time. No stale data.
• 🔍 Semantic Re-Ranking & Synthesis: Instead of relying on static, outdated Vector DBs, the latest JSON metadata chunks are fed directly into massive LLM context windows (Llama-3), allowing the model to dynamically filter noise and synthesize pure insight.
• 🔒 Air-Gapped Privacy Options: Seamlessly switch between the ultra-fast Groq Cloud API and locally hosted Ollama node inference for environments demanding strict HIPAA/GDPR data sovereignty.

---

🏛️ System Architecture

CuraLink operates on a highly decoupled MERN + Aggregator Architecture:

1. Frontend (React, Zustand): A modern, mobile-responsive SPA offering dynamic workspace layouts, sliding off-canvas menus, and robust state persistence.
2. Backend (Node.js, Express, MongoDB): A secure orchestration layer managing JWT authentication, session handling, and stateless API workers.
3. Retrieval Orchestrator: Independent worker services that ping diverse medical APIs (ClinicalTrials v2, PubMed eUtils, OpenAlex works) and normalize their disparate schemas.
4. Inference Layer (LLM): Real-time markdown streaming, inline citation generation, and complex medical reasoning driven exclusively by open-source foundation models.

---

🛠️ Tech Stack

Client

• Framework: React + Vite
• Routing: React Router v6
• State Management: Zustand (with persist middleware)
• Styling: Vanilla CSS + CSS Variables (Deep responsive architecture)

Server

• Environment: Node.js + Express
• Database: MongoDB + Mongoose
• Authentication: JWT (JSON Web Tokens) + bcryptjs
• API Ingestion: Axios (PubMed, OpenAlex, ClinicalTrials.gov)

AI & Inference

• Primary Engine: Groq API (Llama-3 8B / 70B via Language Processing Units for sub-second latency)
• Local Fallback: Ollama (Seamless localhost failover for intranet deployments)

---

⚙️ Installation & Setup

Prerequisites

• Node.js (v18+)
• MongoDB (Local or Atlas Atlas cluster)
• Groq API Key (Free tier available)

1. Clone the Repository

git clone https://github.com/yourusername/CuraLink.git
cd CuraLink

2. Install Dependencies

Install dependencies for both the client and the server.

# Install server dependencies
cd server
npm install

# Install client dependencies
cd ../client
npm install

3. Environment Variables

Create a .env file in the /server directory and configure the following:

PORT=5000
MONGODB_URI=your_mongodb_connection_string
JWT_SECRET=generate_a_strong_random_secret_string
GROQ_API_KEY=your_groq_api_key

# Optional (Defaults to localhost if enabled in UI)
OLLAMA_BASE_URL=http://localhost:11434

4. Run the Application

You can run the full stack concurrently using standard dev commands:

# In the /server directory:
npm run dev

# In the /client directory:
npm run dev

Visit http://localhost:5173 in your browser.

---

🌐 Deployment Recommendations

For quick, hackathon-ready deployments with zero maintenance:

1. Database: Use MongoDB Atlas (M0 Sandbox) and whitelist 0.0.0.0/0.
2. Backend: Deploy the /server directory to Render.com. Ensure your .env variables are securely applied in the Render dashboard.
3. Frontend: Deploy the /client directory to Vercel. Set an environment variable in Vercel (VITE_API_URL) pointing directly to your live Render backend URL.

(Note: Ensure your Settings Page uses the Groq setting for cloud deployment, as local Ollama nodes cannot be accessed by public cloud servers unless securely tunneled via utilities like Ngrok).

---

📄 License

This project was built for the AI Medical Research Assistant Hackathon. Code is open-sourced under the MIT License.