My mother graded papers at the kitchen table every night.

Same chair, same red pen, long after dinner had gone cold. She was a teacher. My dad’s parents were teachers too. That profession runs three generations deep in my blood, and with it comes a very specific kind of knowledge: what it looks like to care more than a system is designed to accommodate.

She never talked about burnout. She just kept going. But I saw what it cost her. That quiet math she was doing in her head every time a student started slipping and there weren’t enough hours left in the week to pull them back. Forty students. One teacher. One hour. The arithmetic never worked, no matter how hard she tried.

That equation is still being run, right now, in classrooms across Malaysia, the Philippines, and Thailand.

The 2024 Southeast Asia Primary Learning Metrics found that 19 percent of Grade 5 students across the region are working at a Grade 3 math level. In the Philippines, by Grade 10, less than two percent of students reach minimum math proficiency. These are not test-score statistics. These are the odds that the next generation can participate in the high-value economy being built around them.

The reason nothing changes is not ignorance. It is the structural bottleneck my mother ran into every night. You cannot personalize instruction at scale with human effort alone. A 1:40 teacher-to-student ratio is not a funding problem. It is a physics problem. The ceiling is the ceiling.

This is why I joined Tupai.ai as Head of Product. And why I put my own capital in alongside that commitment.

Andre and Edmond, the founders, are children of teachers too. When we sat down together, I didn’t need to explain the kitchen table. They already knew it. What they built is a serious answer to a structural problem: an AI tutoring platform that sits alongside the teacher, not instead of her. One that gives each of those forty students a personalized guide while the teacher does what only a human can.

My job is to make sure the product works for the actual human in front of it. A technically sound platform that loses a distracted teenager in the first three minutes is not a solution. Getting the AI right is the starting point, not the finish line. What comes after that is a design problem: how do you build something that a student who already believes she is bad at math will actually trust enough to try?

That is the work I know how to do. And it is the work that needs doing.

Generative AI has changed the underlying economics in a way that matters here. Large language models do not require armies of educators to pre-author every possible learning scenario. The barrier to building something that reflects Malaysian curricula, that speaks in the language a Filipino student actually uses, has fallen in a way it never has before. A Brookings meta-review published earlier this year found that AI tutoring systems can now match the effectiveness of one-on-one human instruction. That was not true five years ago.

The second-order effect is the one most people miss. Private tutoring has always been a wealth signal. Families who could afford personalized instruction gave their children a measurable advantage. Families who couldn’t, didn’t. What we are building does not just improve learning outcomes. It breaks the mechanism by which education quietly compounds inequality across generations.

My mother never got that kind of help.

I’m glad we are finally building it.

Happy Mother’s Day, Mum.