My son had been complaining that the maths at his school in Spain was too easy, which got me thinking about building an app to challenge him. I was also conscious that the Spanish and English curriculums are different, and if we ever moved back to the UK, I wanted to make sure he could keep up with Key Stage 3 maths.
Rather than finding another educational app, I thought it would be a fun project to build something myself. I wanted to keep the idea simple and provide him with a few challenges he could answer each day. Following his love of video games I decided to add some gamification with points, streaks and a bit of competition to keep him interested. It was an opportunity to solve a personal problem, experiment with AI-assisted development and build something useful without turning it into a huge project.
Product screens and details
Today's challenge
Ten questions a day, the same for everyone. The timer starts when the first question appears, with hints and Spanish translations of key terms.
Feedback
Right or wrong, every question ends with a worked solution and a breakdown of how the points were scored.
Select a screen to enlarge it. The screens show a sample run.
The story behind the project
The product
KS3 Maths Daily is a small, mobile-first website offering up to ten maths challenges a day. Each challenge starts when the player presses Start, so simply opening the page doesn't begin the timer. The questions cover Key Stage 3 topics, including arithmetic, algebra, geometry, fractions, probability and statistics. The aim is to make regular practice feel more like a short game than extra homework.
I wanted to gamify the experience without making it childish. Correct answers earn points, with bonuses for speed and solving questions without hints or wrong attempts. There are three progressive hints, daily participation and correct-answer streaks, personal records and a running score. When someone doesn't know how to answer, they can choose 'I don't know this', see the worked solution and move on without being made to feel bad about it.
Behind the scenes, the application keeps track of skills that come easily and those that need more practice, adjusting difficulty and bringing topics back later. There's a separate parent dashboard showing performance by topic and skill, as well as a way to exclude topics his school hasn't yet taught. That distinction matters: not knowing a topic because it hasn't been covered is different from finding it difficult. Questions are in English, with Spanish translations of relevant mathematical terms.
The question bank contains 5,883 generated questions across 162 skills. Rather than asking an AI model to invent a new question every time, the application uses code-based templates to generate questions, answers, hints and worked solutions in advance. It runs on a Cloudflare Worker with D1 for questions and progress. The site can be tried without an account, while sign-in supports saved progress and a separate parent view. A question-reporting feature provides a way to flag errors in the bank.
Building it with AI
I had the idea in mind for a while, then talked it through with ChatGPT on a hike defining and refining all of my ideas into solid product decisions. I wanted a few questions a day, scoring that rewarded speed and independent answers, optional hints, streaks, and a way to identify gaps without discouraging him. I also chose a mobile-first design because a phone is a more natural way for him to pick up and play for a few minutes than sitting down at a laptop.
I used AI to help turn those decisions into a working application rather than spending weeks specifying and building something that might never get used. The first version ran locally in Node.js, using SQLite, and the same core game logic was structured to work in a Cloudflare Worker with D1. The interface is deliberately small and only includes Home, Challenge and Result, with a separate parent dashboard.
The question bank was a more interesting problem than the interface. I didn't want live AI calls deciding whether an answer was correct or inventing inconsistent questions. Instead, I used generators with defined rules so questions, answers and worked solutions could be created in batches and checked. I also wanted the timer, scoring, hints and daily limit enforced by the server, rather than relying on values sent from the browser.
There were still plenty of product decisions to make as it came together, particularly around the scoring penalties, difficulty progression and how to deal with unfamiliar topics. But this wasn't intended to be an elaborate engineering project. It was an opportunity to use AI-assisted development to turn a small idea into something tangible, quickly, and have a bit of fun with it.
Results and reflections
I built a working version, put it online and connected it to the daily email from our Family Assistant so there's a reminder and an easy way to find it. It still needs real use to establish whether the questions are pitched correctly, whether the gamification is motivating and whether the question bank is sufficiently accurate.
For me, part of the value of this project is simply that not every idea needs to become a business, a complex platform or a production-grade system. Sometimes there's a specific problem at home, an opportunity to make something useful and a good reason to build a small tool rather than look for an off-the-shelf product. AI makes those experiments much more accessible, provided I still make the decisions about what to build and why.
Continue exploring
Have an idea or a process you want working?
I take ideas from spec to a working product, and the same approach applies to automation and AI projects inside organisations.



