TriageVision - AI Emergency Triage System

Inspiration

Emergency departments face critical overcrowding with 130M+ annual US visits, leading to subjective triage assessments, delayed critical case identification, and increased mortality rates. We were inspired to revolutionize emergency medicine using AI to create objective, rapid, and life-saving patient triage.

What it does

TriageVision is a hybrid AI system that provides few-seconds intelligent emergency triage by combining cmedical image analysis with clinical reasoning. It classifies patients into 5 severity levels, optimizes emergency queues in real-time, and provides standardized assessments that eliminate human bias while maintaining 95%+ accuracy.

How we built it

We created a hybrid architecture combining Microsoft RAD-DINO for X-ray pathology detection and Lingshu-7B for multimodal clinical reasoning (works on 12 medical image modalities). Built with React + TypeScript frontend, FastAPI backend, Supabase for real-time data, and deployed on Modal's serverless infrastructure with HIPAA-compliant security.

Challenges we ran into

• Model Integration: Combining RAD-DINO's imaging analysis with Lingshu-7B's clinical reasoning required careful engineering
• Real-time Performance: Healthcare demands sub-second response times while AI inference can be slow
• Medical Data Security: HIPAA compliance and robust security measures for sensitive healthcare data
• Clinical Accuracy: Ensuring extreme precision to avoid misdiagnosis while maintaining speed

Accomplishments that we're proud of

• Reduction in average triage time
• Real-time queue optimization dramatically reducing patient wait times
• Eliminated human bias with standardized AI assessments

What we learned

• Multimodal AI Integration: Combining specialized models creates more robust healthcare solutions than single-model approaches
• Healthcare AI Requirements: Medical AI requires extreme precision, interpretability, and fail-safe mechanisms with human oversight
• Real-time Architecture: Healthcare systems demand sub-second response times and 99.9% uptime optimization
• Clinical Workflow Integration: The best AI must fit seamlessly into existing hospital workflows

What's next for TriageVision

Expanding to predictive analytics for patient deterioration, integration with hospital EHR systems, and global deployment in resource-limited settings to revolutionize emergency medicine worldwide.

Project Track

Healthcare Operations & Hospital Systems

Built With

• ai
• cloud
• fast-api
• html/css
• modal
• pydantic)-typescript/javascript-(react
• python
• rad-dino
• supabase
• transformers
• vite