A educational research expedition
across Nepal.

There is a quiet violence in arriving somewhere with the solution already written down. It collapses what is actually there into what you brought with you. It cannot see what does not fit.

The discipline of this project is to delay that collapse. To stay, and to look, and to be told things that change the questions before any product, framework, or recommendation is allowed near them.

Technology will play a role in the future of education in Nepal, of course it will. AI will too. The question is not whether but how, and the only honest source of that how is the room itself: the teacher, the student, the chalkboard, the light coming through the door.

That source has already changed my answer once. I came expecting the gap to be tools. The gap is a connection that holds. What arrives over it — a person, a model, both — matters enormously. It just arrives second.

The first room was to be Crystal Mountain School in Dho Tarap, Dolpa: 4,200 metres, one of the highest schools in the world, several days' walk past where the road gives up. The hardest case first, deliberately — if an idea about education survives that classroom, it will survive most of them.

I did not get there. I walked in as far as Phoksundo and turned back: the coordination the visit depended on had not held at the other end, and a school is not a view you arrive at regardless. The route was mine to plan; being expected was not. I had treated them as one thing — which is the move this page is against, made on the way to the school where I meant to prove otherwise.

So the first thing written down here is not a classroom. It is the distance between a place you can reach and a room that is ready for you. Every school after Dolpa is easier to reach and harder to excuse — and now, also, harder to assume.

The work resumes after Tihar, in the Far-West, unchanged: to write down what is there.

Once the classroom is understood, the tool question becomes honest — and it answers in a different order than I expected.

First, the connection. Not a metaphor for access: an actual link that works on a Tuesday morning in the rain. Many schools have a signal of some kind. Fewer have one that holds long enough for a lesson. What a school already has, what a usable connection would cost it, and whether it stays up are among the plainest things this fieldwork can establish — and nothing further down this list exists without it.

Then the screen. New computers are too expensive for most community schools. Second-hand computers and old mobile phones — the kind already circulating in Nepal's informal markets — are not. An old Android phone connected to a screen, keyboard, and mouse becomes a functional computer. A used laptop from a city reseller costs a fraction of a new one. A device here is not for running anything clever. It is for receiving a class.

Then a teacher at the other end. This is the part I had not understood. A village school is rarely short of teachers in general; it is short of a teacher for that subject — the science, the English, the mathematics nobody on staff was trained in. Those teachers exist. They are in Kathmandu, in Pokhara, abroad, with an hour a week to give. And full-time government teachers, already on payroll, can reach more than the one classroom they stand in. An hour a week, though, is an hour a week.

And a teacher that never leaves. This is where AI earns its place, and I think it earns a large one. A student with a connection already has ChatGPT, Gemini and DeepSeek, free, in a browser — patient at eleven at night, willing to be asked the same question six times, able to explain in Nepali and then in English, and never embarrassed by the asking. For a child whose school has no physics teacher at all, that is not a supplement to a teacher. Some weeks it is the teacher. The volunteer gives the hour that sets the direction; the model covers the many hours in between, and the gap it closes is the largest of anything described on this page.

The order still matters, and it is the whole argument: the models were never the scarce thing. They are free already. What is scarce is the connection and the screen that reach them. Fix those, and a school gets both teachers at once.

Then the thing that makes it repeat. One guest lecture does not do much. The same class every Tuesday for a year does. Matching the hours a volunteer has to the hours a school can use them is ordinary scheduling work, and it gets harder with every school added. So the plan is to write that software: subjects a school needs, hours a teacher has, a timetable both sides can rely on.

Nothing here is new, and the list is short: a link, a screen, an hour, and a model that is already free. The order is the only part worth arguing about.

For all the reasons above, the biggest thing this project needs is not readers. It is teachers — and if you know a subject well enough to teach it for an hour a week over a video call, you are the scarce resource here.

This is not live yet — there is no form, and no class to join tomorrow. What helps now is knowing who is willing and what they can teach, so the first schedule has names in it. Write, and say the subject.