triageX

try out the website at https://triagex.vercel.app ! We spent $ on AI & it was fun.

Inspiration

Singapore's polyclinics face wait times > 3 hrs, draining resources on manual repetitive questioning by nurses. We asked, "How might we turn this passive waiting room into an active autonomous diagnosis stage".

What it does

TriageX is an autonomous voice-first kiosk that interviews patients while they wait, cross-validating real-time symptoms with their FHIR medical history. It uses a multi-agent ensemble to validate diagnosis prints a physical clinical ticket for the doctor.

How we built it

Next.js frontend with ElevenLabs Scribe for voice interaction. The backend orchestrates a parallel "Safety-First" consensus engine using Gemini, GPT, & Llama (via Groq), integrated directly with a Docker HAPI FHIR server.

Challenges we ran into

Mitigating LLM hallucinations & accuracy, so we developed a weighted consensus algorithm. We also had to balance ensemble latency with UX, using progressive rendering & async higlighting.

Accomplishments that we're proud of

We tried our best to mimic HealthHub with a standard-compliant FHIR API handler over H2 database. We also designed a HUD that visualizes the AI's thinking process.

What we learned

We learned that context is king, analyzing symptoms against historical FHIR transforms a simple chatbot into a medical tool. We also balanced it with individual results, most times the patient's symptoms aren't related to their medical history, that's fine too.