Skip to content
missumai
Steal this idea· 8 min read

Steal this idea: WeeNote won Korean school counsellors by keeping their notes off the cloud

A former teacher built a paperwork tool for school counsellors on Replit, watched it fall flat, and learned why: nobody wanted counselling notes leaving the building. He rebuilt it as a desktop app with Claude Code, and payments came in on launch day. Here's the pattern and how to copy it.

about, in March 2026 revenue, founder-reported from memory (Korean won, not converted)
₩20M
schools using it each month, just under 100 of them paying, founder-reported (Apr 2026)
~200
of the 20+ counsellors he interviewed liked it but didn't want their notes in the cloud
70–80%
roughly, and three sleepless days on Replit for the first prototype
$400

WeeNote (위노트) is a desktop program for the counsellors who work in Korean schools' Wee Class counselling rooms. It keeps their counselling logs, turns them into the monthly statistics the principal wants, and produces the upload for NEIS, the national school records system.⁠[1] Its founder, Min Sang-gi, says counsellors tell him work that took two or three hours now takes about ten minutes.⁠[1]

He taught in elementary schools for about 14 years, then worked as a developer for about a year, and started building his own products after watching what AI was turning into. He runs two: WeeNote and a service called IndiePub.⁠[1] The interesting part of WeeNote isn't the features. It's where the data lives.

The pattern in one sentence

If your buyers handle data they would never want to leave the building, don't argue them out of that fear with encryption diagrams: keep the data on their own computer, make that the headline, and build everything else (backups, exports, buying paperwork) around it.

How he found it: one fan, then a flat room

It started on Threads. A school counsellor posted that she was struggling with this exact paperwork, had paid an Excel contractor to build her something, found it too complicated to use, and asked for help. He messaged her and got on a video call; the fix didn't look technically hard, and he could see it would help people straight away.⁠[1]

Claude Code and Cursor weren't part of his routine yet, so he used Replit, because it took care of the database and deployment. By his memory it took about three days without sleep and around $400 to get a prototype.⁠[1] She loved it and sent a long list of requests. He built all of them. Then she showed it to her colleagues, and the mood in their next meeting was flat. The reaction, as she relayed it, was roughly "do we really need this?"⁠[1]

Two things had gone wrong. First, a counsellor who pays an Excel contractor out of her own pocket is, in his words, in the top 1% of the field: her features were real, but most counsellors would never use them. It had become over-engineering.⁠[1] Second, and bigger, people hated the idea of counselling logs being stored in the cloud. Some of what counsellors hear is so sensitive it is never written into official records at all. He explained client-held keys and that the server could never decrypt anything; it didn't help. The objection was the moment the log travels over the network.⁠[1]

“I thought solving the problem meant building the features. But what people were actually worried about was something else.”
Min Sang-gi, in the interview (our translation from Korean)⁠[1]

He didn't jump straight to a rewrite. He kept interviewing, more than twenty counsellors, using the first version. About 70 to 80% said the service was good but storing this information in the cloud was a burden. Several also remembered an earlier, government-led program of the same kind that hadn't taken off because of the same cloud issue.⁠[1] So version two went local.

What happened after it went local

When counsellors asked whether it was safe, he could now say that all data is stored only on the teacher's computer, encrypted and backed up. He says nobody raised a security concern after that, and payments kept coming in from launch.⁠[1]

  1. 2025

    Leaves his developer job early in the year to build his own products.⁠[1]
  2. Before launch

    Threads post, video call, and a Replit prototype in about three days for about $400. Flat reaction from colleagues; 20+ interviews follow.⁠[1]
  3. Late Dec 2025

    Opens payments by bank transfer only, with no card payment gateway, because approval takes two weeks to a month. A payment arrives the same day, another a few days later.⁠[1]
  4. Early Jan 2026

    Version two launches. Schools' budget year starts in March, but a few counsellors scrape together leftover budget anyway.⁠[1]
  5. Late Feb–Mar 2026

    Adds a card payment gateway before the March contracts; the official launch runs from late February into early March. Payment alerts pour into his Telegram.⁠[1]
  6. Mar 2026

    About ₩20 million in revenue, he thinks (he says he doesn't remember exactly); his Toss Payments limit of ₩10 million a month was exceeded soon after opening (founder-reported).⁠[1]
  7. Apr 2026

    About 200 schools using it monthly, just under 100 paying (founder-reported). Interview published 20 April.⁠[1]

Pricing is a yearly licence: Basic ₩300,000, Pro ₩480,000 and Premium ₩960,000. He says Basic and Pro each make up about half of sales, Pro brings in a bit more money, and Premium sells now and then.⁠[1] Free users come in through trials of one to three months, and he watches how many keep using it as their trial ends.⁠[1]

How it spread inside schools

Each school normally has one Wee Class counsellor, so he assumed one licence per school. Then a school bought two: the career-guidance teacher wanted it too. Another bought four, one for the counsellor and three for homeroom teachers, after the counsellor offered it on the school messenger to anyone who wanted to buy along with her. He rewrote the homepage to speak to counsellors, career teachers and homeroom teachers, not just counsellors.⁠[1] Today's homepage does exactly that.⁠[2]

Local storage also helped with the gatekeepers. A district education office called him, sceptical, asking about data security: schools in its area kept asking whether they could use WeeNote. After hearing that data stays local, encrypted and backed up, the office told him it saw no problem and would tell schools so.⁠[1] WeeNote is also listed on EduZip, the Ministry of Education's edtech information site, which he says keeps school approval reviews smooth.⁠[1]⁠[2]

What local-first looks like on the page

WeeNote's homepage, read on 14 November 2026, sells the architecture as much as the features (our translation):

  • The promise up top: data is "stored safely only on the school computer, not online".⁠[2]
  • Cloud vs local, side by side: a comparison table lists the cloud risks (data on outside servers, mass leaks if a server is hacked, no access if the vendor closes, data locked to the vendor) against WeeNote's local approach.⁠[2]
  • Specifics, not adjectives: the database and backup files are encrypted with AES-256, server traffic is only licence checks and updates over SSL/TLS, and automatic backups run on a schedule.⁠[2]
  • Your data stays yours: if the licence expires, a viewer still opens the records and exports them to Excel or PDF.⁠[2]
  • The AI stays local too: its built-in AI drafts logs and reports and runs only on the computer.⁠[2]
  • An analogy buyers already trust: typing student data into Excel doesn't give it to Microsoft, and typing it into WeeNote doesn't send it to WeeNote.⁠[2]

Notice how buying is designed for a school office too: pay with the school's admin card and the licence key, delivery note, spec sheet and card slip arrive by email right away, with a quote and bank transfer option for schools that need it.⁠[2]

The stack, as described

First versionReplit, chosen because it handled the database and deployment.⁠[1]
Desktop appElectron with SQLite, a file-based database. Human developers he asked were mostly against it; he debated it with Claude Code, which said it was widely used in commercial apps, and went ahead.⁠[1]
Coding todayEverything goes through Claude Code, run straight from a terminal. He dropped Cursor (he never opened the files anyway), tried Conductor and the spec tool OpenSpec, and dropped both as too much process for a solo early product. A few Claude skills hold the things that won't change; the rest is prompts.⁠[1]
ShippingNo branches and no test code: build, test by hand, ship before 8 a.m. because teachers arrive at 8:30.⁠[1]
SecurityAES-256 encrypted database and backups on the teacher's PC; SSL/TLS for licence and update traffic.⁠[2]
PaymentsBank transfer at first, then Toss Payments; school admin card checkout today.⁠[1]⁠[2]

How to build your own

The steps follow what WeeNote did. Where a step is our own suggestion, it says so.

  1. Look for professionals stuck in Excel

    He says wherever people are doing a job in Excel, there's an opportunity, sometimes even just a good template.⁠[1] His first user had literally paid an Excel contractor.⁠[1]
  2. Prototype fast, then show it to the average user

    A fan who brings you the problem is often the most advanced user in the field. Build the prototype for her, then put it in front of ordinary colleagues before adding her wish list.⁠[1]
  3. Ask where the data is allowed to go

    In his 20+ interviews the blocker wasn't a feature, it was the cloud.⁠[1] Ask it directly, and ask whether a similar product failed before; his market remembered one that did.⁠[1]
  4. Go local, and accept the work that comes with it

    He weighed installs, database migrations that now run on each customer's machine, and backups people will blame you for if they fail.⁠[1] Ship automatic encrypted backups from day one.⁠[2]
  5. Keep the server for licences and updates only

    That is the only traffic WeeNote describes, and it is what lets the homepage say no counselling data is sent.⁠[2]
  6. Lead with the architecture, in plain words

    Headline promise, a cloud-vs-local table, named encryption, and an everyday analogy.⁠[2]
  7. Remove buying friction for institutions

    Card checkout with paperwork emailed instantly, a quote option, an FAQ that answers the approval question, and a listing on the official registry.⁠[2] Time it for the buyer's budget year.⁠[1]
  8. Answer support like a person, fast

    He reports four or five bug reports a day once customers arrived in March, answered around the clock, fixed in 30 minutes to an hour with AI, then explained back to the reporter. Nobody asked for a refund over a bug, he says.⁠[1]
  9. Build the part that can't be copied

    He thinks someone could clone the features in about a month. His answer is a community of more than 1,000 school counsellors with about 20 mentor teachers, and member-only material for verified WeeNote users.⁠[1]

The twist: he took "AI" out of the name

The product was first called WeeNote AI, to show off the on-device AI he was proud of. In an in-person interview, he asked counsellors who hadn't adopted it why not, and many said it was because the name had AI in it. Explaining that the AI never touches the cloud didn't matter: in counselling, many see AI involvement as out of place, and a product with AI in its name as one whose quality they can't trust.⁠[1] He calls it the moment he realised he was still a developer caught up in the technology.⁠[1] The AI is still there, as one feature that runs locally; it's just not the pitch.⁠[2]

Where else this could work

Our suggestions, not tested by WeeNote: any job where people keep confidential notes about other people and do the reporting by hand. Think private therapists, school nurses, social workers or HR case files. The test is the same one he ran: ask twenty of them where they'd accept their notes being stored, before you build the sync server.

Risks before you copy it

  • Desktop is harder to run. Installs, migrations on every customer machine and backup blame were the reasons he hesitated.⁠[1]
  • Bugs land on you directly. Four or five a day at the start, even after his own testing and a hired QA tester.⁠[1] Losing data is the one bug you can't afford; he draws the line there too.⁠[1]
  • Approval rules are the buyer's call. WeeNote's FAQ explains why it thinks schools don't need a committee review, but also says each school may judge differently.⁠[2] Don't promise what a regulator hasn't said.
  • Budget calendars. Schools buy from March; launching in January meant waiting for most of them.⁠[1]
  • Features are easy to clone. His own estimate is about a month, which is why he spends heavily on the community.⁠[1]
  • "AI" can scare the buyer. In a sensitive field, the word itself cost him users.⁠[1]

What to take from this

  1. 01

    Where data lives is a feature

    The cloud, not a missing button, was the blocker.⁠[1]
  2. 02

    Don't build for the top 1%

    His first user's wish list was over-engineering.⁠[1]
  3. 03

    Interview before you rewrite

    20+ conversations decided the architecture.⁠[1]
  4. 04

    Make buying boring

    Card checkout and instant paperwork for school offices.⁠[2]
  5. 05

    Hide the tech in the pitch

    Dropping "AI" from the name removed a reason to say no.⁠[1]
  6. 06

    Own a community

    Features copy in a month; 1,000 colleagues don't.⁠[1]

For more apps that began on Replit, see the best apps built with Replit, where WeeNote is one of the picks. For another tool that grew from one specific audience, read how FootballGPT built an AI assistant for the people its founder already taught.

All stories

The newsletter

More stories like this, by email

Occasional email with new projects on the leaderboard and new build stories. Every number sourced, like the site.

We'll email a link to confirm. Unsubscribe in one click.

Sources

  1. [1]혼자 AI로 만든 SaaS, 월 2,000만 원 찍었습니다 (“A SaaS I built alone with AI hit ₩20 million a month”)하조은 (Ha Jo-eun) on YouTube, an interview with WeeNote's founder Min Sang-gi · Apr 20, 2026 · checked Nov 14, 2026In Korean. We read the video's uploaded Korean subtitles in full on 2026-11-14; every English line on this page is our own translation or paraphrase. Useful timestamps: 2:59 the Threads post, 4:10 the Replit prototype, 4:52 the flat reaction, 7:36 the interviews, 8:54 pricing, 9:34 the first payments, 11:07 March revenue, 12:03 schools using it, 14:20 the education office call, 15:33 Claude Code, 19:04 bugs and support, 21:47 the community, 25:53 the name. Everything he says is founder-reported; no dashboard is shown.
  2. [2]위노트 - 학교상담시스템 (WeeNote homepage)WeeNote · checked Nov 14, 2026In Korean, read on 2026-11-14 (our translation). These are the company's own claims about its product; we have not audited the software.
  3. [3]WeeNote (위노트): ₩20M/mo in MarchProvenStartups · checked Nov 13, 2026A summary of the same interview (captured 2026-07-25). Read on 2026-11-13; we checked every point we use against the interview itself.

Something wrong or out of date? Tell us and we'll fix it.