CasperCoach

In March 2025, while still in high school, I built CasperCoach, an AI test-prep platform for the CASPer exam. I built every layer myself, scaled it from zero to 5,700+ verified medical school applicants, and it has generated $12K+ in revenue.

This is the story of how it happened, and how it actually works under the hood.

What CASPer is

CASPer stands for Computer-Based Assessment for Sampling Personal Characteristics. It is a situational judgment test used by over 500 medical, dental, and health professional programs. Instead of testing what you know, it tests how you think: ethics, empathy, communication, professionalism. It has quietly become one of the most important non-academic factors in admissions decisions.

Why I built it

I had to take CASPer for my application to Western Engineering. When I went looking for ways to practice, I only found two kinds of resources. Random YouTube videos about how the test works, and courses sitting behind $200 paywalls selling their self-proclaimed frameworks.

Neither one let you actually practice. The videos were surface level. The courses charged hundreds of dollars to basically tell you to be empathetic and structure your answer in three parts. There were no realistic scenarios, no way to write under timed pressure on prompts you had not seen, and no feedback on whether your answer was any good.

So while studying for my own test, I started building the thing I wished existed.

What I built

The core is an AI scoring engine on the OpenAI API that grades open-ended answers in real time. I wrote custom prompts and a rubric-based evaluation so the feedback lines up with how the real test is actually marked, not just generic praise. Every attempt is tracked, so you can see your scores improve over time instead of guessing whether you are getting better.

Around that I built AI-generated scenarios that match the format and difficulty of the real exam, a clean interface that mimics the actual testing environment so practice feels like the real thing, and guides that show you why certain response patterns score well and others tank your score.

Making it feel fast and worth using

A prep tool is useless if people bounce after thirty seconds. I spent real time on the experience: sub-second page loads and an interface designed to keep you in a focused practice loop. The result is a 15-minute average session, which is roughly six times the 2.3-minute web average. People actually sit down and practice.

Payments

The whole thing is self-serve. I built a Stripe subscription system with webhooks that instantly unlock Pro access the moment a payment goes through, no manual steps on my end. That system is what turned CasperCoach from a project into something that has made $12K+ in revenue while I sleep.

Keeping the free tier honest

Once there was a paywall, people started farming free trials with duplicate accounts. So I built an anti-fraud layer using email normalization and fuzzy matching to catch the duplicates and stop the abuse. I later spun that idea out into its own machine learning project.

Getting users

To get the first wave of users I leaned on organic content. Almost every day I wrote long, in-depth CASPer guides on Reddit, then dropped the link at the bottom for anyone who wanted to practice what they had just read. The posts took a while to write because I focused on genuinely useful content built to spread and earn credibility.

These Reddit posts would often get 10 to 50k views each:

From there growth took off. Pretty quickly I reached a few hundred signups who started actually practicing.

About four months later, in August, I hit 1,000 users.

Where it is now

Over a year later I have scaled it to several thousand users, and it helps applicants get through their test every week. You can see it live at caspercoach.org.