Built for my own kids first.

The problem was not effort

My boys were not lazy and they were not struggling for want of trying. They were sitting in classrooms moving at the pace of thirty children, quietly accumulating gaps. Once a gap opens in maths it compounds — every new concept leans on the one underneath it, and a child who missed the one underneath stops being able to follow.

The obvious answer is a tutor. But a good one-to-one tutor in New Zealand is expensive, and the harder problem is finding one whose schedule, location and approach actually fit your family. Plenty of families cannot afford it. Plenty who can still cannot find one.

What I wanted instead

Something patient enough to explain the same thing a fourth time without a sigh. Something that remembered what my son found hard last week, rather than starting fresh every session. And something that would tell me the truth about how he was going, in plain language, without me having to hover or quiz him at dinner.

Most of what I looked at was a bank of exercises with a progress bar bolted on. That is not a tutor. A tutor is a relationship that accumulates.

Why it took so long to start building

I spent months on the architecture before writing the product, which is an unusual way to start and I think the right one here. The hard parts of a tutor for children are not the teaching prompts. They are the boundaries: what the system is allowed to remember, what it is allowed to change on its own, what counts as evidence that a child has actually learned something, and what happens when it is not sure.

Those boundaries are much harder to add later than to design in. So Abbi separates the teaching from the memory, the memory from the safety rules, and the tone a child hears from the curriculum truth underneath it. A model does the explaining, inside a system that decides what it may do.

Where it is up to

Abbi is being built deliberately narrow to begin with — one curriculum, one subject, one stage — because proving the whole loop properly on a small slice is worth more than a broad demo that works in a screenshot. It grows from there.