Clinical MultiAgent CDSS

Developed for the Google AI Agents Capstone Project

🔬 Research Foundation

This project is a production-ready implementation of clinical research focused on diabetes management. As a first-author researcher published in Springer Nature, I developed this system to bridge the gap between static machine learning models and active clinical reasoning.

• Original Paper: [https://doi.org/10.1007/s13410-026-01658-3]
• Core Methodology: Bridging XGBoost predictive ensembles with grounded medical guidelines (ADA 2023 / PES 2022).

🏗️ System Architecture

The system uses a four-agent pipeline orchestrated via the Google Agent Development Kit (ADK) and an MCP Server bridge.

The Agents:

1. Clinical Validator (The Gatekeeper): Handles data extraction, PII masking, and physiological bounds checking.
2. Risk Predictor: Executes a custom XGBoost ensemble via a standardized MCP tool.
3. Guideline RAG: Retrieves medical diagnostic thresholds from a specialized Markdown knowledge base.
4. Clinical Auditor (Chief Resident): Cross-references model output with medical truth to identify discrepancies.

🛡️ Safety and Security

• Emergency Bypass: Immediate hospital referral triggers if glucose > 400 mg/dL or age < 21.
• Privacy: Automatic PII filtering at the validation stage.
• Logic Integrity: The Auditor flags model "hallucinations" if they contradict established medical guidelines.

🚀 How to Run

1. Clone this repository.
2. Create a .env file with your GOOGLE_API_KEY and HF_TOKEN.
3. Install dependencies: pip install -r requirements.txt
4. Run the UI: streamlit run app.py

📺 Demo Video

Watch the full architectural walkthrough and live clinical simulation run on YouTube:

Clinical CDSS Architecture Demo

Click the image above to view the demonstration video.