Clarity AI

Title Slide
Why it Matters
Tech Stack
Chat Interface
Hospital Map Page

Clarity AI – Project Story for DevPost

(Trauma-smart hospital navigation powered by multimodal AI)

💡 Inspiration

Talking with ER and trauma nurses, we kept hearing the same pain points:

Trauma is still a top preventable killer.*
Over-loaded hospitals miss golden-hour care.*
A bad first stop triggers a costly transfer.*
We asked: “What if every patient walked into the right hospital first try?”

🩺 What it does

Patient input – Snap a photo, speak symptoms, or type a quick note.
Multimodal triage – ChatGPT Vision + text LLM predicts urgency & required services.
Smart matching – Google Places finds nearby hospitals → we fuse with the AHA dataset (23 k + rows) in MySQL and a live-load simulator.
Ranking engine – Scores each hospital on distance, open capacity, trauma level, imaging units, etc.
Instant guidance – Front-end map (React/Vite + shadcn/ui) shows the best route and sends an advance-arrival bundle to the ER.

🛠 Tech stack

Layer	Tech
ETL / Data	Python pandas ➜ Docker-hosted MySQL (American Hospital Association 2023)
Back-end	Node 18, Express, Google Places API, OpenAI Multi-modal
Front-end	React + Vite, shadcn/ui, Leaflet map
Infra	Docker Compose for DB & API containers

🚀 Hackathon features shipped

Google Places ↔ AHA fuzzy matcher (name + geo radius)
Capacity simulator — seeds live‐load columns for every hospital
Hospital-scoring service — weights capacity, ED, trauma level, MRI/CT, burn beds…
Interactive map — user taps “Find care” → ranked pins & one-click navigation

🧩 Toughest challenge

Integrating two noisy, totally different datasets (unstructured Google JSON + rigid AHA schema) inside Docker-MySQL and surfacing them fast enough for a real-time query. Lots of longitude/latitude headaches, name normalisation, and SQL tweaking before we hit sub-100 ms responses.

📚 What we learned

ETL discipline pays off—renaming 250+ AHA columns to sane snake_case unlocked faster dev.
Even simulated capacity data changes the ranking dramatically; patients rarely need the physically closest hospital.
Clear separation (repo 🗄 → service 🤖 → controller 🚦) kept the Express code maintainable under crunch time.

🔮 What’s next

EHR / HL7 FHIR polling for real bed & ICU load.
Provider dashboard—shows inbound Clarity AI cases with ETA and auto-populated notes.
HIPAA compliance pipeline—end-to-end TLS, zero-retention PHI, audit logging.
Insurance & ambulance hooks—optimize cost and transport mode.