Agents MD

Watching a loved one battle a disease is heartbreaking.

Watching them struggle in an overcrowded ER, waiting for care, is devastating.

Across the country, emergency rooms are in crisis—overcrowded, understaffed, and overwhelmed. Nurses face impossible workloads, making life-or-death decisions under extreme pressure, while patients endure agonizing delays in time-critical care.

Agents MD is transforming ER triage by addressing these urgent challenges. Our multi-agentic AI approach brings together competing AI models, each with specialized diagnostic expertise, to refine differential diagnoses in real time. By reducing diagnostic uncertainty and easing the burden on overworked nurses, we help accelerate patient care, enhance accuracy, and mirror the collaborative decision-making of medical teams—ensuring no patient is left waiting when every second counts.

Our Solution

Agents MD leverages the power of large language models in a collaborative framework that mimics how medical professionals work together to reach consensus on patient care priorities. By combining multiple specialized AI agents, we create a system that is:

• More accurate than single-model approaches
• Transparent in its decision-making process
• Supportive of existing medical workflows
• Responsive to time-critical situations

Installation

Prerequisites

• Python 3.12.7
• pip 25.0
• Anaconda or Miniconda
• Git

Setup Steps

1. Clone the repository

   git clone <repository-url>
   cd AgentsMD

2. Set up Conda environment

   conda create -n agents-md python=3.12.7
   conda activate agents-md

3. Install dependencies

   pip install -r requirements.txt

4. Environment Setup

   • We provide a template dotEnv file with the required API keys structure
   • Create your own .env file by copying the template:

     cp dotEnv .env

   • Get your API keys from:
     • OpenAI - For AI models and diagnosis generation
     • AssemblyAI - For speech-to-text transcription
   • Replace the placeholder values in .env with your actual API keys:

     OPENAI_API_KEY="your-openai-api-key"
     ASSEMBLYAI_API_KEY="your-assemblyai-api-key"
     

Running the Demo

1. Navigate to the demo directory

   cd demo

2. Start the application

   python app.py

3. Access the web interface

   • Open your web browser
   • Go to http://127.0.0.1:5000 (the default Flask development server)

Features

• Real-time speech-to-text transcription
• AI-powered triage assessment
• Differential diagnosis generation
• Case discussion and detailed output viewing
• Recording and transcription capabilities
• Patient prioritization system

Important Notes

• Ensure all required API keys are properly set in your .env file
• The application creates directories for storing recordings, transcriptions, and other data
• The system uses SQLite for storing conversation history
• Make sure you have sufficient disk space for storing audio recordings and transcriptions

Repository History

This repository represents a significant evolution from its original foundations. Initially inspired by AgentLaboratory, which provides an end-to-end autonomous research workflow using LLM agents, we saw an opportunity to adapt this multi-agent approach to address critical challenges in emergency medical care.

The entire project was conceptualized and implemented in less than 24 hours during Philly Codefest, a hackathon hosted by Drexel University's College of Computing & Informatics. This intensive development sprint demonstrated the potential of adapting agentic AI systems to tackle critical real-world challenges, such as emergency room triage optimization.

The journey of this codebase involved:

1. Initial fork of AgentLaboratory into AgentsMD for medical applications
2. Collaborative development to transform the codebase for ER triage use cases during the hackathon
3. Creation of this new repository due to substantial divergence from the original codebase, including:
   • Complete refocus on medical diagnosis and triage
   • Integration of medical-specific LLM agents
   • Addition of speech-to-text capabilities for patient interactions
   • Implementation of real-time diagnostic systems
   • Development of a web-based interface for medical staff

While we maintain the spirit of multi-agent collaboration from AgentLaboratory, this project has evolved into a specialized medical tool with its own unique architecture and purpose and hence, has been moved to a separate new repository.

Acknowledgments

This project was developed in collaboration with Shailesh Sridhar and builds upon work done in the AgentsMD repository, which was originally forked from AgentLaboratory. We express our gratitude to both repositories for their contributions to this project's development.