🏥 TriageFlow

AI-assisted patient intake + staff-controlled triage + live waiting queue

TriageFlow collects patient intake before staff interaction and uses AI to convert it into a clean, structured summary that clinicians can review fast.

🛑 AI never makes the final decision. Every urgency level and triage outcome is confirmed by staff.

---

Getting started (Website)

https://triage-flow-nine.vercel.app/

---

⚙️ Getting started (local)

1) Install dependencies

npm install

2) Start Postgres (Docker)

docker-compose up -d

(Optional) Check the container is running

docker ps

3) Run the app

npm run dev

Open: http://localhost:3000

---

🚀 Why TriageFlow

Triage desks lose time re-asking the same intake questions, especially during busy hours. TriageFlow speeds up the process by:

• capturing the story once
• turning it into a consistent format
• letting staff review, override, and finalize in seconds
• keeping patients informed with live queue status and wait estimates

---

✅ What it does

🧑‍🦱 Patient experience

• 🖥️ Open the site on any device
• 📝 Complete a short guided intake (designed for under 2 minutes)
• ⏳ Enter a single waiting queue
• 📍 See:
  • current status (Submitted -> In Review -> Finalized -> Ready)
  • estimated wait time
• ✅ Get marked Ready when staff is ready to see them

👩‍⚕️ Staff experience

• 📥 Live intake feed of incoming submissions
• 🚦 Automatically sorted by:
  1. urgency
  2. arrival time
• 🔎 Review:
  • full intake transcript
  • AI-generated structured JSON
  • suggested urgency and red flags
• ✍️ Staff can:
  • override urgency
  • add short notes
  • finalize triage
  • mark a patient as Ready
• 🔄 Queue updates immediately (Polling or SSE)

---

🧠 AI output (strict JSON, validated server-side)

Every intake produces a validated structured JSON response including:

• 🚨 suggestedUrgency: Low | Medium | High | Emergent
• 🤕 symptoms: extracted symptom list
• 🕒 duration: onset + duration (when available)
• 📈 pain: pain score or description (when relevant)
• 🧯 redFlags: key risk indicators
• 🩺 staffSummary: concise clinical summary
• 🙂 patientSummary: simple plain-language summary

✅ The AI output is treated as a suggestion and is always shown alongside staff controls.

---

🧾 Waiting queue + estimated wait time

TriageFlow uses one queue to keep the demo clear and easy to understand.

Ordering rules

Patients are ordered by:

1. ✅ Final urgency (confirmed by staff)
2. 🕒 Arrival time

Wait time estimate

Estimated wait is calculated as:

• patients ahead × average service time

Service time is based on a simple service category estimate (preset averages).
When the queue changes, estimates update automatically.

---

🔁 Realtime updates

TriageFlow supports realtime via:

• ✅ polling (simple and reliable)
• ✅ SSE (fast, low overhead)

Either is acceptable for demo. The goal is that judges can see updates happen live.

---

🧱 Tech stack

• 🖼️ Frontend: Next.js (App Router) + TypeScript + Tailwind CSS
• 🧩 Backend: Next.js API Routes
• 🗄️ Database: PostgreSQL
• 🔁 Realtime: Polling or Server-Sent Events (SSE)
• 🤖 AI: LLM API with strict JSON output + validation
• 🚀 Deploy: Vercel + managed Postgres

---

📸 Screens to look at (judge-friendly)

• Patient Intake: guided intake form + confirmation
• Patient Status: queue position + estimated wait
• Staff Dashboard: live feed + review panel + actions

---

🛡️ Safety and scope

TriageFlow is an intake and workflow tool. It does not:

• diagnose
• recommend treatment
• replace clinical judgment

It supports clinicians by organizing information and making review faster.