AegisHealth-AI

A production-ready FastAPI backend designed to streamline emergency room operations using Advanced Gemini Models. Developed for the NEXORA Global Hackathon.

The Main Problem

Emergency Rooms frequently suffer from overcrowding and delayed patient assessments, which can lead to critical misprioritization. Traditional triage processes rely on manual evaluations that are susceptible to human error and delays, especially when dealing with a high influx of patients. There is a pressing need for a rapid, accurate, and autonomous risk assessment system to assist medical staff.

Execution

AegisHealth AI serves as an autonomous triage agent leveraging the Manchester Triage System guidelines.

• AI Integration: Evaluates patient vitals and intake details using the gemini-3.5-flash model for high-speed clinical risk assessments.
• Multi-Modal Vision Triage: Incorporates medical image analysis alongside vital signs to enhance the triage accuracy via advanced vision models.
• Real-time Telemetry: Utilizes WebSockets to simulate and stream live patient vital signs seamlessly.
• Robust Data Handling: Implements Async SQLAlchemy with an SQLite database to reliably store patient data, vital signs, and triage assessments.

Result

The system successfully ingests patient data to immediately output a structured JSON evaluation.

• Patients are accurately categorized into strict urgency levels (RED, ORANGE, YELLOW, GREEN, BLUE) with an associated risk score from 1 to 5.
• The AI provides detailed clinical reasoning, a top-3 differential diagnosis, and immediate recommended ER protocols to initiate.
• A live queue API helps medical personnel monitor waiting patients ordered by urgency and time in the queue.

🛠️ Tech Stack in Healthcare

• Backend Framework: FastAPI, WebSockets
• AI & Machine Learning: Google GenAI SDK (gemini-3.5-flash, gemini-3.0-flash)
• Database & ORM: SQLite, Async SQLAlchemy
• Data Validation: Pydantic models
• Frontend Integration: Jinja2Templates, HTML/CSS

Built with ❤️ for the NEXORA Global Hackathon

Built With

• css
• html
• jinja
• python
• sqlite
• websockets