Inspiration
Medical diagnosis is a high-stakes, complex process where a single physician's perspective can be limited by cognitive biases or incomplete information. We were inspired by the collaborative "tumor board" or Multi-Disciplinary Team (MDT) meetings in hospitals, where specialists from different fields (like radiology, oncology, and pathology) come together to review complex cases.
Our idea was to create a digital version of this process: a multi-agent system where different AI "specialists" could analyze a patient's case, debate the potential diagnoses, and challenge each other's assumptions. We named it Socrates to reflect this Socratic method of using structured debate to uncover the most accurate, well-reasoned conclusion.
What it does
Socrates is a full-stack clinical decision support tool.
Patient Creation: A user (a physician) first creates a new patient case, inputting their name, age, gender, height, weight, and a detailed medical description, including symptoms and history.
Debate Configuration: The user then selects a "panel" of AI specialist agents (e.g., Cardiology, Neurology, Respiratory) to consult on the case.
Live Debate: The system initiates a "debate" where the selected agents autonomously analyze the patient's data. They propose initial hypotheses, challenge each other's findings, and work toward a consensus, all in a real-time, viewable transcript.
Summary & Insights: Once the debate is complete, the user is presented with a final summary page showing the most likely diagnoses, the confidence score from each agent, and the full debate log, providing a transparent and explainable (XAI) trail of reasoning.
How we built it
Socrates is a full-stack application with a clear separation between the agent backend and the user-facing frontend.
Backend: We used FastAPI (Python) for its high-performance, asynchronous capabilities. This was essential for managing the multiple, concurrent "thinking" processes of the AI agents without blocking the server.
Frontend: We built a dynamic and responsive user interface using React and TypeScript.
Database: We used PostgreSQL with the Prisma ORM to store patient information and the transcripts of every debate.
Log in or sign up for Devpost to join the conversation.