Inspiration

I had to book a medical appointment recently and it was annoying. I realized if it's this frustrating for me as an English speaker with free time during business hours, it must be absolutely brutal for people who don't speak English or Spanish, or who can't call during the day because they're working.

What it does

Medilin lets patients fill out a healthcare intake form in their native language, then an AI agent calls the hospital and books the appointment for them, and you get a confirmation email in your language.

How we built it

  • Frontend: Next.js and React with Tailwind for the intake form and UI
  • AI Voice: ElevenLabs for the conversational AI agents
  • AI Language Model: Google Gemini for translation and conversation logic
  • Email: Resend for sending appointment confirmations
  • Deployment: Vercel

Challenges we ran into

The biggest pain point was getting the ElevenLabs agent preview to work the way I wanted. I initially wanted the two AI agents (patient-side and hospital-side) to actually speak to each other with live audio output, but the audio streams were completely out of sync and the whole thing fell apart. Had to pivot to just showing transcripts instead.

Accomplishments that we're proud of

First time using ElevenLabs and honestly, it's incredible. The voice quality and conversational ability is way better than I expected. Even in the simulated version, you can see how natural the agent negotiation feels. We also nailed the multilingual flow. The entire system works in the patient's native language from intake form to confirmation email, and the translation quality with Gemini is solid.

What we learned

ElevenLabs is a game-changer for voice AI. This was my first time using it and I'm impressed by how easy it is to create natural-sounding conversational agents. The API is clean, the voice quality is production-ready, and the agent capabilities are far beyond basic text-to-speech.

What's next for Medilin

  • Implement actual phone calling integrating ElevenLabs' telephony features to make real calls to clinics instead of simulating responses
  • Calendar integration connecting to patient calendars for better appointment negotiation
  • HIPAA compliance ensuring all patient data handling meets healthcare regulations
  • Partner with a few healthcare providers to test the system with actual appointment workflows

Built With

Share this project:

Updates