Education
Funding
Bootstrap
My Role
Co-founder
Timeline
June 2023 - Now

Medical teacher in WhatsApp

The challenge

For the past 20 years, my mother has been active in medical education. She built from scratch a school, named FormaScience that supports 1st year medicine students in France pass their exams. Medical studies are known to be some of the hardest studies in the world. And their methodology, whatever they do special, seems to work, as over 85% of their students get the exam despite an average acceptance rate in universities below 20%.

Over time, their school developed an e-learning platform that replicates the quality of an in-person class, but at home. It even became part of the curriculum for in-person students. They break down the program into learning objectives, each objective into interactive classes with drawings, MCQ, explanatory videos... But when students has a very specific question, they gave a couple of options:

  • search on Google, find a PDF or a medical article, and search for the part that answers their question
  • go on the e-learning platform, find the topic and then the class that addressed this question
  • ask the question on a forum that FormaScience opened

The solution

So when ChatGPT came out in 2022, I got curious. Does it have sufficient knowledge to answer the typical questions of a 1st year medicine student? And while we're at it, can it become a full learning assistant that can also explain concepts in a pedagogical way and ask MCQs in the way a professor would?

There are a couple of requirements though:

  • The interface would need to be faster than anything they currently use to find information.
  • It needs to be accurate, so there needs to be some kind of supervision / benchmarking of the information provided by specialists
  • The initial prompt needs to embed all of the knowledge that FormaScience gained in the last 20 years

I really believe in the fact that new solutions should be developed in a user-centric way. Do people need yet another app, yet another website? Most likely, no. So building a chatbot / voicebot on WhatsApp became an obvious choice. Everybody's on WhatsApp. Chances are that these students spend a couple of minutes / hours a day talking to friends on WhatsApp. But it also has the advantage of not being a real social network, in the sense that there is no external source of distraction on the app.

So the concept was born: build a chatbot, on WhatsApp, with my mother, that would take her knowledge, feed it to ChatGPT, and answer all the questions that students might have. To help them learn and succeed their first year medicine studies. medprof.ai was born.

Building

I'm not a really good programmer, but I wanted to build this from scratch. Here's the stack:

  • Memberstack + Webflow for sign-up
  • Render.com to deploy the webhook & the functions
  • Twilio for the WhatsApp agent
  • Google Cloud for the storage

I built it in 15 days on the side of my job:

  • 10'000 students
  • 550'000 conversations handled
  • 1200 queries per day on average

We're doing the acquisition via TikTok mostly: https://www.tiktok.com/@medprofs

This is what launch adoption was like with TikTok:

Number of users on medprof.ai

What I learned

  • Build things made for scale & automation technically
  • Simple > Complex
  • Distribution > Idea
  • TikTok + Emailing automations
  • Better websites