Your AI Copilot for Emergency Triage
Continuous re-assessment. Age-aware. Transparent.
The nurse always has the final say.
The Problem
Emergency departments across India face systemic overload. Static triage scores decay the moment they're assigned.
Illustrative
How It Works
Four steps from arrival to continuous prioritization.
Arrive
Walk in, approach the kiosk, choose your language — Hindi, English, or both.
Speak
Tell MediPilot what's wrong. It extracts structure — it never decides acuity.
Assess
Vitals measured, age-aware thresholds applied, red flags caught in under a second.
Prioritize
Continuous re-scoring every 60 seconds. The sickest patient is always seen first.
Built for Real Hospitals
Every design decision answers a question a nurse, a patient, or a judge would actually ask.
The Human Lane
Prefer a person? The nurse sits beside you, opens the same system on her tablet, and walks through the questions together. The technology adapts to the patient — never the other way around.
Everything at a Glance
Three explainability channels on every patient card. What drove this score, what was considered but didn't move it, and what was said verbatim. Every factor visible, every override recorded.
Same Number, Different Meaning
38.5°C in a 3-year-old and a 75-year-old are not the same finding. Six age strata, each with independently calibrated thresholds. The model that ignores age gets one of them wrong and looks confident doing it.
Red Flags — Instant, Deterministic
Eight hardcoded rules. Active labour, altered consciousness, chest pain with radiation, uncontrolled bleeding. If any fires, the patient bypasses the model entirely and goes straight to Red. No AI involved. No delay.
Surge Mode — What Stretches, What Holds
At 3× normal load, Green and Yellow cadences stretch to buy time. Red cadences never stretch. The system refuses to relax what cannot be relaxed — and says so explicitly on screen.
The Nurse Closes Every Loop
One tap to override any recommendation. A 16-field legal record captures exactly why. The system's job is to surface the right patient — the human's job is to decide what happens next.
Under the Hood
A 10-layer pipeline from voice to risk score.
The language model extracts structure.
It is grammar-constrained to never produce a diagnosis or acuity level.
That's enforced by the output schema, not a preference.
System Constraint
No diagnosis.
No acuity band.
No "could this patient die" field.
Grammar rejects any output outside the allowed schema.
The Five Screens
Each screen runs on a different device in a real ED. Click to explore.
Nurse Board
The queue — three time facts per card; breaches glow
Patient Intake
Four-step branch, typed tree, voice — in Hindi or English
Waiting Hall
Public display — token numbers only, no names, no acuity
Judge Control
R slider, surge rate, simulation speed, scenario jump
Audit Ledger
Override records rendered verbatim, trust panel, model card