Why I built CleverSprogs

As my son worked through Key Stage 1, 2 and on towards the 11+, my wife and I both took an active interest in his education.

However we found private tuition expensive, but also hit and miss in terms of quality and communication, so we found it difficult to intervene in a timely manner where necessary to help him.

So I built the platform I wished I'd had, using AI-assisted app development to do it on my own.

CleverSprogs is an online learning and assessment platform for UK children in Years 1 to 6, covering maths and English plus 11+ preparation. The intention is to continue into KS3 and beyond, in the near future.

I wanted two things from a learning platform: good value for money - imperative in the current economic climate - and a clear view of a child's strengths and weaknesses.

Each day a child gets a short round of questions pitched at the topics they're ready for. When they get one wrong, the app offers a step-by-step worked solution, highlighting the step where they went wrong.

Behind the scenes, we track which topics are secure and where they are struggling, and we tell the parent where their child needs a hand.

Monthly reports show the mistakes a child keeps making, side by side with the correct working, and parents can print worksheets built from the exact questions their child got stuck on.

It's in final testing now, ahead of opening to families. It took me under six months. This is the story behind it, and what it means if you're a founder thinking about commissioning software.

How fast is "fast"?

I measured what went into CleverSprogs and estimated how much work the same thing would take a conventional team:

  • Software: about 280,000 lines of working code. At 100 to 200 finished lines per developer per day, that's 6 to 13 person-years.
  • Questions: nearly 40,000 of them. At 8 to 15 minutes each to write and check, that's 3 to 6 person-years.
  • Teaching guides: 344 printable guides for parents. At half a day to a day each, that's about 1 to 2 person-years.

Pooled across a five-person team, that comes to between two and four years. I did it in five and a half months.

Estimated months to build CleverSprogs: one engineer with AI 5.5 months; five-person team 25 to 49 months

These figures are my own estimate, and the rates behind them vary from team to team. Many of the questions are built from rules rather than written out one by one, so I quote the cautious end: four and a half times faster than a small team. On the generous figures, it's closer to nine.

An experienced engineer checks everything the AI writes

AI tools wrote a large share of the code. I decided what to build, how the pieces fit together, and whether each piece was right before it went in.

In an earlier post we covered how AI-built apps go wrong when nobody with engineering experience is checking the output. The defence is a set of automated tests that check the software works as intended.

CleverSprogs has about 260,000 lines of them, almost as much as the app itself. When the AI changes something and breaks something else, a test fails and I fix it before it reaches anyone.

Reliable where it matters

A general-purpose AI chatbot can tutor a child, but it can also get the maths wrong with total confidence. AI researchers call this a hallucination. A busy parent may not spot it, and a seven-year-old won't.

CleverSprogs takes a different route. Every question a child sees comes either from a question bank written in advance or from fixed rules written in code. No question is made up by AI while a child is answering, and every answer is marked by code that follows the same rules each time.

AI helped me write the questions faster, and I checked them, or the rules that build them, before any child could see them. This is where bespoke software earns its keep: you decide exactly how it behaves, and it behaves that way every time.

What this means for your app

The same approach that built CleverSprogs works for client projects. A senior engineer working with AI can deliver in months what used to take a team years, and that brings bespoke software within reach of founders and sole traders with smaller budgets.

You get software built around how your business works, checked by someone who knows what good engineering looks like.

If you have an idea you'd like to turn into an app, get in touch for a free, no-obligation chat.

Sources