DiagScore

DiagScore is an AI-assisted clinical consultation platform for live doctor workflows. It combines a React dashboard, a FastAPI backend, and a LiveKit voice agent to capture consultation data, track checklist coverage, and surface decision-support signals during a patient encounter.

Architecture Overview

• Frontend: React + TypeScript dashboard for intake, live consultation, vitals, notes, ECG upload, and AI-assisted views
• Backend: FastAPI service for authentication, session management, consultation state, scoring, and storage integration
• Agent: LiveKit voice agent for real-time listening, note capture, and checklist guidance
• Data layer: MongoDB, Redis, Qdrant, and MinIO

Tech Stack

• React, TypeScript, Vite, TanStack Query
• FastAPI, Pydantic, Beanie, Motor
• MongoDB, Redis, Qdrant, MinIO
• LiveKit

How It Works

1. A clinician logs in and creates a consultation session.

2. The frontend requests a LiveKit token and joins a room.

3. A voice agent joins the same room and listens to the consultation.

4. The backend stores consultation state and returns updated views to the dashboard.

5. The UI shows checklist progress, likely profiles, recommended tests, notes, and risk summaries.

Key Features

• Consultation session management
• Real-time voice assistant support
• Clinical checklist tracking
• Suggested tests and profile scoring
• Vitals and ECG capture
• Consultation notes and synthesis panels

Running the Project

1. Clone the repository.
2. Create an environment file with the required API keys and service settings.
3. Start the stack with Docker Compose or run each service separately.
4. Open the frontend and backend API locally.
5. Sign in, create a consultation, and start a LiveKit session.