MediVoice 🎙ī¸

Speak your symptoms. Get your MC. Skip the queue.

Agora Voice AI Hackathon ¡ Singapore 2026

😷 Problem

2.3M polyclinic visits/year in Singapore — half for minor ailments.

  • Doctors spend 30–40 min per patient on a 3-min clinical decision
  • Patients take half-days off just to get an MC
  • Waiting rooms = unnecessary exposure risk
  • Services locked out of non-English speakers

The bottleneck is data collection, not medical judgement.

💡 The Insight: One Doctor. 100x the Reach.

Traditional telehealth just moves the clinic online. We scale it.

Traditional Clinic Telehealth App MediVoice
Patients per doctor/hr ~2 ~4 ~80
Clinic branches served 1 1 Unlimited
Data collection Doctor Doctor AI Agent
Languages 1–2 1–2 4

One AI doctor. Every branch. Every language. 24/7.

Why Agora? ConvoAI runs STT → LLM → TTS inside the call, sub-200ms via SD-RTN. Not duct-taped together.

✅ Solution: AI Telehealth Platform

❌ Before ✅ MediVoice
Patient wait 2–3 hours at clinic ~10 min from home
Doctor time 30–40 min per case < 2 min async review
Data collection Doctor does it manually AI conducts full interview
MC issuance In-person only PDF delivered instantly
Language English only EN ¡ 中文 ¡ BM ¡ āŽ¤āŽŽāŽŋāŽ´ā¯
Throughput ~8 cases/hr per doctor ~80 cases/hr per doctor

One doctor. 10x the impact. Zero waiting room.

🛠ī¸ Tech Stack

Layer Technology
Voice Transport Agora RTC SDK (agora-rtc-sdk-ng)
AI Agent Agora Conversational AI Engine (agora-agent-client-toolkit)
LLM OpenAI GPT-4o
TTS ElevenLabs / Cartesia
Backend Python ¡ FastAPI ¡ Uvicorn
Auth & Token Agora Token Builder ¡ python-dotenv
Database Couchbase
PDF Generation ReportLab
Frontend React 19 ¡ TailwindCSS ¡ Vite
UI Components Radix UI ¡ Lucide React ¡ shadcn/ui

Agora handles the full loop — STT → LLM → TTS — over low-latency SD-RTN. Not a wrapper.

🏗ī¸ Architecture

Layer Component Tech Role
🧑 Patient React App Agora RTC SDK Voice input · live transcript
☁ī¸ Agora Cloud SD-RTN ¡ ConvoAI STT → GPT-4o → TTS AI clinical interview
⚙ī¸ Backend API Server FastAPI ¡ Couchbase Token gen ¡ agent control ¡ case DB
đŸŠē Doctor Dashboard React ¡ TailwindCSS Review ¡ approve ¡ escalate

All real-time intelligence flows through Agora — no custom STT or TTS pipeline.

🎤 Patient Flow

  1. Open MediVoice → Start Consultation
  2. AI agent interviews: complaint, duration, severity, symptoms, meds
  3. Red-flag detected? → agent stops, advises 995 / A&E + immediately escalates to on-duty doctor via live Agora RTC call
  4. Summary generated → patient confirms → enters doctor queue

No app. No typing. No waiting room. Just speak.

đŸŠē Doctor Flow

Doctor opens dashboard → sees AI-structured case summary.

Decision Action
✅ Approve Select MC duration (1–3 days) → PDF issued
📝 Modify Edit diagnosis or duration
🚨 Escalate Flag for in-person, patient notified

One screen. One click. Under 2 minutes.

🧑‍⚕ī¸
HITL — Human In The Loop
AI never issues an MC autonomously. Every case requires an explicit doctor decision before anything is sent to the patient.

📈 Market & Affordability

Metric Value
SEA Telemedicine Market (2025) $2B+
SG polyclinic visits / year ~2.3M
MC-eligible cases / year ~1.1M

Healthcare shouldn't cost more than a meal.

Traditional Clinic MediVoice
Patient cost $15–30 + transport $8.99
Time spent Half a day ~10 min
Where Queue at clinic From home

🍗 The price of a chicken rice. Healthcare is affordable again.

🏁 The Ask

Partner with one polyclinic group for a 3-month pilot.

  • MC wait: 2.5 hrs → 30 min
  • Doctor throughput: 10x
  • Built on Agora RTC + ConvoAI ¡ FastAPI ¡ React