I went to a Korean school conference β the kind where teachers from Korean weekend schools come together to share what's working in their classrooms. A good chunk of it was about AI: how to use it to plan lessons, make materials, generate practice, save time on the endless small tasks that fill a teacher's week.
The content was genuinely excellent. People had done real work and were sharing it generously. I left with pages of notes and a nagging feeling I couldn't name for a couple of days.
Eventually I worked out what it was. Almost none of it was going to get used.
The ideas were good. The problem was the shape of the whole thing.
Every technique lived somewhere else. This one needs that site. That one needs a different tool. Another needs a third thing, plus an account, plus a bit of setup. And a lot of them needed a paid subscription β paid by the individual teacher, out of pocket, with their own card.
Put that in front of a Korean school teacher and look at it honestly. Many are volunteers or near-volunteers. They teach on Saturdays, often around a full-time job and a family. Now the ask is: sign up for five services, learn five interfaces, remember which one does which, keep five subscriptions alive on your own money, and do that maintenance forever, for a class you might be teaching out of love rather than salary.
A handful of the most determined teachers will do it. Everyone else will go home inspired, get through the next few weeks, and quietly keep doing things the old way β not from lack of interest, but because the cost of adoption was set higher than the job could carry.
The gap wasn't knowledge. The knowledge had been transmitted perfectly; that conference did its job. The gap was that the knowledge arrived in a form that required each teacher to individually rebuild the same setup, alone, and pay for it.
That's a structural problem, and structural problems don't get solved by more good presentations. You can hold that conference every year, and every year the same teachers will adopt and the same majority won't, because the obstacle was never the information.
Which means the useful thing to build isn't another AI tool or another set of tips. It's the missing layer underneath β somewhere all of this already lives, already set up, already paid for, so a teacher's first step is "log in" instead of "assemble a toolkit."
Hangeulmaru β νκΈλ§λ£¨ β is one place with everything a Korean school needs: AI help built specifically for Korean school teaching, plus the ordinary running of a school β lesson planning, classrooms, the administrative work that eats weekends.
Two decisions shape it, and both come straight out of that conference.
This is the part I care most about. A school signs up; its teachers just use it. Nobody enters a personal card, nobody expenses anything, nobody maintains a subscription for a volunteer job.
The moment a tool requires a teacher to spend their own money, it has stopped being a tool for that school and become a tool for the few most motivated individuals in it. Moving the decision up one level β from person to institution β changes who can realistically participate from "the determined few" to "everyone."
General-purpose AI can do a lot, but it doesn't know what a Korean weekend school is. It doesn't know these students are mostly heritage learners whose Korean varies wildly within one classroom, that the school year runs on Saturdays, or what a Korean school teacher's materials actually look like.
Teachers end up supplying all that context by hand, every single time, in every prompt. Baking it into the tool removes an invisible tax that was being paid over and over by people who were already short on time.
One school is using it. That's the honest status.
I'd rather say that plainly than describe a platform. One real school with real teachers and real Saturdays is worth more to me at this stage than ten schools signed up out of curiosity β because everything that's wrong with it right now, I'm going to hear about from people who are actually depending on it.
More schools. Korean schools solve the same problems separately, everywhere, with no shared infrastructure. Multiple schools on one platform means the work of setting something up once can serve all of them.
Parents, too. This is the part I'm most interested in. Right now Hangeulmaru is for teachers, but a Korean school runs on parents as much as staff β driving on Saturdays, practicing at home, trying to keep a language alive in a house where the kids answer in English. They're doing the hardest and least supported part, usually with no visibility into what's happening in class.
A version where parents can see what their child is learning and get help continuing it at home is where this stops being a teacher tool and starts being a home for the whole school. That's the direction β not built yet, but it's what the shape is for.
If you teach at a Korean school, run one, or are a parent at one β I'd genuinely like to hear from you. Get in touch; that's how this gets built right.
More origin stories and bugs in the build log, or on Instagram.