EPCID- Early Pediatric Critical Illness Detection

Home Screen
Login Screen
Dashboard
My Children
Care guide
Doctor Reports
Dosage Calculator
Family Sharing
AI Assistant
Find Care
Growth Chart
Health and Guide
Mental Wellness
Light Mode
AI Engine
Learn More
PEWS Assessment

Inspiration

Parents often struggle to decide when a sick child needs urgent care. Pediatric illness can deteriorate quickly because children compensate until they suddenly crash. Warning signs often appear hours before a crisis but go unnoticed or misunderstood.

I built EPCID to act as a real-time pediatric triage assistant. The goal is simple: help parents recognize early warning signals and guide them toward the right care decision using evidence-based clinical logic and multimodal AI.

What it does

EPCID is a real-time pediatric illness detection and triage platform powered by multimodal AI.

Parents describe symptoms through voice or text, enter vital signs, and show visible symptoms through the camera. The system analyzes the information and produces structured clinical guidance.

The platform provides:

Risk classification from Low → Moderate → High → Critical
Age-aware interpretation of pediatric vital signs
Possible clinical causes based on symptom patterns
Safe care guidance and monitoring instructions
Clear escalation advice such as home monitoring, pediatric consultation, urgent care, or emergency services

The system combines conversational interaction with structured clinical scoring so parents receive understandable guidance backed by clinical reasoning.

How I built it

EPCID uses a cloud-native architecture designed for real-time medical triage.

Backend

FastAPI microservices deployed on Google Cloud Run
OAuth2 authentication and secure API endpoints
AI orchestration layer with clinical scoring modules
Structured logging and observability

Frontend

Next.js progressive web application
Parent dashboard for symptoms and vital tracking
Real-time triage results and quick action recommendations

AI

Gemini 2.5 Flash via Vertex AI
Structured JSON outputs for reliable clinical analysis
Safety prompts and guardrails to enforce conservative medical advice

Data Layer

Streaming ingestion of pediatric vitals:
- Temperature
- Heart rate
- Respiratory rate
- Oxygen saturation
Explainable symptom knowledge base

Challenges I ran into

Maintaining low latency when calling cloud AI models
Ensuring consistent structured JSON responses from the model
Designing prompts that keep medical advice safe and conservative
Balancing explainability with real-time responsiveness

Accomplishments

Built a fully deployed end-to-end prototype
Delivered real-time pediatric risk scoring
Implemented structured clinical reasoning outputs
Designed a cloud-native architecture ready for wearable integrations
Created a live demo dashboard running on Google Cloud

What I learned

Healthcare AI requires a safety-first engineering mindset.

Models must produce grounded outputs. Systems must log every decision and allow human review. Conservative escalation rules protect users when uncertainty exists.

Reliable medical AI depends on transparency, guardrails, and explainable reasoning.

Risk scoring formula

EPCID computes a weighted clinical risk score using multiple physiological indicators such as fever, heart rate, respiratory rate, and oxygen saturation.

[ Risk = \sum_{i=1}^{n} w_i \cdot s_i ]

Where: