MediScan-AI

Inspiration

Rural clinics lack specialists, time, and reliable internet Patient triage decisions must be made quickly and safely Privacy-preserving AI can support clinicians without cloud dependency

What it does

Performs local-only AI triage using symptoms, images, and cough audio Classifies patient urgency (Low / Medium / High) Stores assessment history locally Checks medication interactions and conditions Provides analytics for clinic-level insights Supports multiple languages for accessibility

How we built it

Frontend: Next.js with a clinician-friendly UI Backend: FastAPI (Python) AI Pipelines: Image processing Audio feature extraction Triage scoring logic Local storage for history and analytics Privacy-first architecture (no cloud, no APIs)

Challenges we ran into

Designing healthcare AI without cloud services Combining multimodal data into clear triage results Maintaining simplicity for real clinic workflows Balancing medical responsibility with AI assistance

Accomplishments that we're proud of

Built a fully local, privacy-first healthcare AI system Implemented multilingual support Added history, medication safety, and analytics modules Designed a system suitable for real rural clinic use

What we learned

Privacy-first AI requires different system design Explainability is critical in healthcare tools UI clarity matters as much as model accuracy Real-world AI is about systems, not just models