Health AI

Inspiration

Accessing healthcare quickly is still frustrating especially considering the long waits, unclear clinic availability, and no easy way to know where to go when symptoms appear. Our inspiration came from a simple question: what if healthcare guidance felt as immediate and intuitive as chatting with an assistant, while automatically guiding you to real nearby clinics?

Health AI was born from the idea of combining AI-driven symptom analysis with real-world location awareness, so users aren’t just told what to do, they’re shown where to go.

What it does

Health AI is an intelligent healthcare assistant that:

Lets users describe symptoms through a conversational AI interface
Uses location access to automatically find nearby walk-in clinics using Google Maps & Places
Interactively highlights and displays clinics as interactive cards and pins on a live map
Maintains conversation history and context across sessions
Uses Eleven Labs Text to Speech to automatically book appointments on the user's behalf.
The result is a seamless flow from symptom → recommendation → real-world action.

How we built it

Frontend: React + HTML + CSS + Tailwind

Custom UI for chat, clinic cards, and Google Maps integration
Real-time map panning, zooming, and marker highlighting
State-driven sync between clinic cards and map pins

Maps & Location:

Google Maps JavaScript API
Google Places API for nearby walk-in clinics
Geolocation API for user location consent

AI & Backend:

Utilizes Microsoft Foundry to fine tune models
Utilizes C#'s Semantic Kernel to allow base models to perform agentic actions on the backend
Uses Supabase/PostgreSQL + EF Core to store user information, such as conversations, messages, symptoms, and appointments.
Utilizes Twilio + Eleven Labs to provide real-life sounding calls that books appointments on the user's behalf
Uses Terraform + JetBrains CI/CD to significantly speed up development speeds

UX Details:

Smooth animations, dark mode, and cursor-based effects
Clinic websites accessible directly from the UI
Clear visual hierarchy between chat, clinics, and map

Challenges we ran into

Synchronizing map behavior with UI state

Ensuring the map didn’t constantly re-center or jitter when clinics were selected required careful control of map bounds and effects.
Handling inconsistent API responses
Some clinics didn’t have websites, open hours, or complete data, we had to design graceful fallbacks.

Avoiding excessive Google Places calls

We optimized searches so the Places API only runs once per user location.
Maintaining smooth UX across async data flows
Chat responses, clinic discovery, and map updates all had to feel instant and connected.

Accomplishments that we're proud of

Fully integrated AI + Maps + real clinic data in a short hackathon timeframe
Seamless interaction between clinic cards and map markers
Clean handling of real-world edge cases (no website, closed clinics, missing data)
A polished, usable product, not just a demo
Most importantly, we built something that feels practical, intuitive, and immediately useful.

What we learned

Designing with real APIs introduces complexity that mock data never shows
Maps UX requires restraint, fewer automatic movements often feel better
Clear state management is critical when multiple UI components depend on the same data
Healthcare tools need clarity and trust more than flashy features