MedX

MedX is a research-oriented framework that leverages Large Language Models (LLMs) to automatically extract temporomandibular joint (TMJ)-related comorbidities from unstructured clinical notes. The pipeline transforms raw patient text into structured summaries and generates an interactive visual dashboard for cohort-level analysis.

This tool supports both clinical research and decision-making by surfacing patient-level insights and population-level statistics such as comorbidity frequencies, means, and standard deviations.

---

🔧 Features

• LLM-based Comorbidity Extraction: Supports BART and DeepSeek-based models
• Automated Clinical Summarization: Generates structured outputs per patient
• Chunked Input Handling: Efficiently processes long clinical documents
• Interactive Dashboard: Visualizes cohort-level statistics and trends

---

📁 Project Structure

• model_run_chunks.py – Run inference using the selected LLM model (BART or DeepSeek)
• model_fine_tune.py – Fine-tune your own LLM on labeled data
• dashboard.py – Generate an interactive Dash-based visualization for cohort summaries
• requirements.txt – List of required Python packages

---

🚀 Getting Started

1. Install dependencies:

   pip install -r requirements.txt

2. Run predictions:

   python model_run_chunks.py
   

3. Fine-tune a model (optional):

   python model_fine_tune.py
   

4. Launch the dashboard:

   python dashboard.py
   

The dashboard visualization is built using Matplotlib