← AI Lab

1 Million Questions Answered. 500 Exams. And AI Is Building Most of It.

November 20, 2026

Tigertest just passed 1,000,000 questions answered. We’ve also reached 500 different exams and qualifications to study for, each with its own bank of practice questions.

What started as a little side project to help me pass my DMV test has turned into something much bigger than I expected.

Weekly active learners

Weekly active learnersWeekly active learners on Tigertest, week ending Oct 2 to Nov 20 2026: 111, 90, 111, then after sign-up became optional 213, 391, 917, 1,664 and 1,845. 111 Oct 2 90 Oct 9 111 Oct 16 213 Oct 23 391 Oct 30 917 Nov 6 1,664 Nov 13 1,845 Nov 20 No sign-up needed
Each bar is the 7 days to that date. Once people could practise without signing up, weekly learners grew 1,560% (17x) in five weeks, from 111 to 1,845.

This week, people visited Tigertest from 130 countries.

In Kyrgyzstan, future doctors cramming for their licensing exams. In Fiji, nurses preparing to work overseas. In Brunei, a new driver learning the rules of the road.

Teachers have started bringing entire classes too. They create a class, share a link, and see which questions their students are getting wrong.

New sign-ups per month

New sign-ups per monthNew Tigertest sign-ups per month, Dec 2025 to Nov 2026: 61, 284, 305, 572, 495, 576, 637, 696, 565, 422, 436, and 1,205 in November so far. 61 Dec 284 Jan 305 Feb 572 Mar 495 Apr 576 May 637 Jun 696 Jul 565 Aug 422 Sep 436 Oct 1,205 Nov
Dropping the sign-up wall didn’t cost sign-ups. November has 1,205 so far (to Nov 20), nearly triple October.

I love seeing something I built being used by people all over the world.

Claude builds Tigertest

Most new exams now go from idea to live without me touching them.

I’ve built an automated workflow of AI agents using Claude Code:

  1. Find demand. One agent looks at what people are searching for on Tigertest.
  2. Pick what to build. Another looks at demand, usage, and which exams people value enough to pay for.
  3. Write the questions. A third builds the question bank, answers, and explanations.
  4. Check the answers. A fourth verifies everything against official sources.

These aren’t prompts I run manually. The agents work together, passing their output to the next stage.

When something goes wrong, I fix the workflow rather than just the individual output.

My job is increasingly just deciding what matters.

Three things I’ve learned

1. Ship the small version.

Tigertest started as one DMV practice test. I built exactly what I wanted: something I could use to practise on my phone in bed, then sit down at my desktop and take a proper exam.

It turned out other people wanted the same thing.

2. Feed the machine.

When AI makes a mistake, don’t just correct it. Feed the learning back into the workflow so it doesn’t happen again.

A lesson that doesn’t make it back into the machine is wasted.

3. Automate the quality checks.

AI makes generating thousands of questions easy. Making sure they’re accurate is harder.

If you’re automating creation, you need to automate verification too. Speed shouldn’t come at the expense of accuracy.

On to the next million

The person who answered question 1,000,000 got Premium for life. Seemed like the least I could do.

Thanks to everyone who’s used Tigertest, shared it, reported a mistake, or suggested something new.

It’s pretty amazing seeing a little project I built for myself helping people prepare for exams all over the world.

On to the next million.

Search for your testSet up a class