CareCall Story

Inspiration

I was struck by global statistics:

  • 970 million people live with a diagnosable mental disorder, yet 70 % receive no support.
  • One in three older adults report loneliness and one in four fall each year and wait hours for help.
    I saw that nearly everyone even isolated seniors can be reached by a simple phone call and decided to build a service that brings AI driven check ins to those who need them most.

What it does

CareCall places daily voice calls or SMS check ins to recipients using AI generated, personalized scripts based on the users condition . If it detects concern it alerts the care taker

How we built it

  • Backend
    • FastAPI running locally with SQLite
    • ngrok to expose local endpoints for Twilio and Vertex AI webhooks
    • Twilio for outbound calls, recordings, and SMS
    • Vertex AI Gemini for script generation, Cloud Text to Speech for voice, and Whisper for transcription
  • Frontend
    • Next.js 14 with App Router
    • Passwordless magic link auth via Resend and HTTP only cookies
    • React Query for data fetching, Tailwind CSS and shadcn/ui for styling
    • Environment variables managed in env

Challenges we ran into

  • Refactoring authentication when switching from Auth0 to Resend magic links
  • Reconfiguring for a fully local development setup with ngrok instead of a public cloud host
  • Phasing out Cloud Run deployment and updating CI/CD pipelines and secret management accordingly
  • Resolving SSR errors by clearly separating Next.js server and client components

Accomplishments that we’re proud of

  • Building a seamless AI pipeline that generates a personalized custom script with follow up, and analyzes sentiment
  • Enabling daily check ins for vulnerable people via phone or landline, bypassing internet or app requirements
  • Achieving reliable end to end development locally using ngrok and FastAPI with real webhooks
  • Delivering a polished Next.js dashboard with passwordless authentication and real time status updates

What we learned

  • Voice first solutions can bridge digital divides and reach isolated populations
  • Proper client and server separation in Next.js 14 prevents build errors and improves maintainability
  • Effective use of ngrok accelerates webhook testing without public hosting overhead
  • Integrating multiple AI services (Vertex AI, TTS, Whisper) requires careful resource naming, region configuration, and IAM permissions

What’s next for CareCall

  • Add multi language support for non English speakers
  • Implement caregiver dashboards for bulk recipient management and analytics
  • Integrate additional alert channels (email, push notifications)
  • Explore predictive AI models to detect early warning signs before crises occur
  • Add conversational AI support so calls can adapt based on freeform user responses
  • Prepare for production grade deployment with a scalable database, managed cloud hosting, and monitoring
  • Host the service in a production environment with CI/CD pipelines, logging, and automated scaling

Built With

Share this project:

Updates