Participation MarkHackRift 2025Track: Smart Nation

The Future of
Autonomous Triage

Addressing the challenge: "How might we design user-centric solutions that improve the quality of life in Singapore?"

The Problem

The "3-Hour" Bottleneck

As Singapore ages, our polyclinics are overwhelmed. The average wait time for a consultation often exceeds 3 hours.

Currently, highly trained nurses spend thousands of hours manually asking:"What are your symptoms?" and "Do you have any drug allergies?".

Data Fragmentation

Doctors often lack immediate context on a patient's history, relying on verbal confirmation which is slow, unreliable, and prone to memory error.

Why TriageX is Better

Traditional Triage
Sequential, Manual, Slow (5-10 mins/pax)
TriageX Kiosk
Parallel, Instant, Scalable (Avg 45s)
Human Judgment
Prone to fatigue & cognitive bias after long shifts.
Ensemble AI
3 Models (Gemini + GPT + Llama) cross-validating every decision.

Technical Execution

Voice Input

ElevenLabs Scribe v1

Consensus Engine

Gemini 2.5 + GPT-OSS + Llama 4

Triage Ticket

P-Score & Notes

Safety-First Consensus Algorithm

We poll three state-of-the-art models via Vercel AI SDK. We prioritize safety over averages.

IF any_model_vote == "P1_CRITICAL"
THEN final_score = "P1" (Override)
ELSE final_score = CEIL(AVG(votes))

Structured Output Enforcement

Unlike standard chatbots, TriageX is the only solution using Zod schema validation to enforce strict JSON outputs. This ensures data is 100% compatible with hospital databases.

{
  "acuity": "P2",
  "actions": ["RICE", "X-Ray"]
}

Kiosk User Guide

1

Select Profile

Simulate scanning an NRIC. The system pulls relevant history (e.g., Diabetes) from the local FHIR database.

2

Click-to-Speak

Tap the microphone once to start describing symptoms. Tap again to stop. No need to hold the button.

3

Receive Ticket

Review the AI's consensus log and "Print" your Triage Slip for the doctor.

FHIR Simulation

We run a local Docker instance of hapiproject/hapi to simulate the National Electronic Health Record (NEHR).

GET /Patient/P001
> 200 OK
> Conditions: [Hypertension, Asthma]
Real-time Context Injection