CareFever

CareFever: Your 24/7 AI-powered fever assistant.

Comment

🩺 CareFever: Your AI-Powered Fever Assistant

💡 Our Inspiration

CareFever was born from a familiar 3 AM panic — a high fever and a flood of anxious Google searches.
Is it just the flu, or something serious like Dengue? Instead of spiraling into uncertainty, we wanted to build a calm, clear, and personalized assistant that helps you understand your symptoms safely.
CareFever is that solution — a one-stop AI platform that gives context-aware, reliable health guidance when you need it most.

⚙️ How We Built It

We designed CareFever as a modern, decoupled web application with a strong focus on scalability, performance, and security.

🖥️ Frontend

  • Built using React + Tailwind CSS
  • Secured with Clerk Authentication
  • Provides an intuitive and responsive user interface

🧠 Backend

The Node.js/Express backend acts as the “brain” of CareFever, orchestrating multiple powerful services:

  • 🤖 Google Gemini API — for advanced AI-powered symptom analysis
  • 🗣️ Vapi.ai — for a real-time, multilingual voice assistant
  • 📞 Twilio API — sends dual-channel SOS alerts (SMS + WhatsApp)
  • 🔥 Firebase Realtime Database — logs and tracks all user reports securely
  • ⚡ Redis Cache — stores scraped health news for instant speed and low latency

🧩 What We Learned

🪄 Prompt Engineering is 90% of AI

Our most significant learning came from finetuning the Gemini prompt.
We discovered that the quality, safety, and empathy of the AI’s response depend entirely on the prompt design.
After dozens of iterations, we built strong guardrails to ensure the AI:

  • Responds safely and responsibly
  • Avoids giving medical advice for severe symptoms
  • Maintains an empathetic and calm tone

🧬 Hybrid AI Architecture

We implemented a hybrid architecture:

  • Gemini runs on the backend for secure, deep analysis
  • Vapi.ai runs on the client for low-latency, real-time voice interaction

This setup offered the perfect balance of speed, safety, and scalability.

🌍 Voice AI is More Than Just Language

We learned that multilingual AI isn’t just translation — it’s about cultural nuance and emotional accuracy.
Each language had its own way of describing symptoms, and we had to ensure that:

  • The AI’s tone remained empathetic
  • Speech-to-text worked accurately across accents
  • The assistant sounded human, not robotic

🚧 Challenges We Faced

1. Finetuning the AI Symptom Checker

Early versions of the Gemini model tried to give direct advice for severe symptoms like chest pain — a serious safety issue.
We spent considerable time refining the prompt with strict red-flag rules to ensure the AI:

  • Recognizes emergencies immediately
  • Escalates to emergency response instead of providing analysis

2. Making the Voice Assistant Truly Multilingual

Creating an accessible, multilingual system was far more than adding translation.
We had to:

  • Manage multiple voice models
  • Ensure accurate transcription for various accents
  • Preserve empathy and clarity across all supported languages

This challenge pushed us to think about AI communication beyond code — making it truly inclusive and human.

🧑‍💻 Tech Stack

Layer Technologies
Frontend React, Tailwind CSS, Clerk
Backend Node.js, Express
AI Google Gemini, Vapi.ai
Database Firebase Realtime Database
Messaging Twilio (SMS + WhatsApp)
Caching Redis

🌟 Team Vision

We believe AI in healthcare should empower — not overwhelm — users.
CareFever combines the empathy of a friend, the speed of AI, and the safety of professional guardrails to make health guidance accessible anytime, anywhere.

"Don’t Google your symptoms. Ask CareFever."

Built With

  • clerk
  • elevenlabs
  • express-database-&-caching:-firebase-(firestore)
  • express.js
  • firebase
  • framer-motion-backend:-node.js
  • framermotion
  • gemini
  • javascript
  • node.js
  • postman
  • react
  • redis
  • redis-authentication:-clerk-ai-apis:-google-gemini-(for-symptom-analysis)
  • tailwind
  • tailwind-css
  • twilio
  • vapi.ai
  • vapi.ai-(for-voice-assistant)-cloud-services-&-apis:-twilio-(for-sms/whatsapp-sos)
Share this project:

Updates