✨ Founder story

How a vibe-coded church app reached $14.4K MRR — and survived its own breach

Anton Karbanovich built live AI sermon translation for his own international church. By October 2026 it was doing $14,404 a month — and along the way he publicly owned a security breach that cost him $2,500 in Stripe fees.

$14.4KMRR (Oct 19, 2026)
100Subscriptions
$31.2KLast 30 days
$196.5KAll-time revenue

A sermon nobody could follow

Anton Karbanovich is a Belarusian founder living in Cyprus, and he did not set out to build a church product for the market. He set out to fix his own Sunday. His international church had visitors from ten countries, and most of them could not follow the sermon. That is the kind of problem that repeats every single week, whether or not anyone is selling a solution for it.

What he built is Glossa: live AI translation for church services and Christian events. The pastor speaks, people in the pews scan a QR code, and each person hears or reads the sermon in their own language on their phone. It covers more than 100 languages with no headsets or extra equipment, handles Bible references and church vocabulary, and can speak in a voice modeled on the pastor’s. The pricing is simple on purpose: $5 an hour per language on pay-as-you-go, monthly plans at $99, $299 and $499, and 4 free hours to try it.

He is not a first-time founder. His LinkedIn lists earlier companies, and he posts often about building. But Glossa is the one he built for a room he was already sitting in — which is why the product fits the room so precisely. A freelancer Dan Raine talked to had turned the church niche down; few tech companies chase churches. A church with members from ten countries, though, has a real weekly need, and nobody else was serving it.

Free first, then Product Hunt

Karbanovich did not start by building. He started by translating — by hand, for free, at his own church’s events. Only after the room had used it did he turn it into a product, and only then did he launch it on Product Hunt, in September 2025. The order matters: the first users were people he already knew, with a problem he had already watched.

The build itself was AI-assisted from early on. In March 2025 he described Glossa as “all built with ai tools and a middle developer.” By 2026 he was building with Claude Code, and after parting ways with his technical co-founder he wrote that he and Claude Code “finished more tasks than I could have imagined.” AI does the work in the product too — speech recognition, translation and the spoken voice — and AI helped build the thing that does the work.

Distribution has been deliberately narrow: replying in founder threads on X, where he keeps sharing the link, and word of mouth between churches. There is no paid acquisition story here. The growth came from a niche that compounds by itself — one church that works becomes the reference for the next church down the road, and every Sunday is another demo.

The breach he posted about

On February 28, 2026, Karbanovich posted something most founders would hide. His “vibe-coded startup was exploited.” His API keys had ended up in the part of the app anyone can see, and someone used them to charge 175 customers $500 each before he could rotate the keys. He lost $2,500 in Stripe fees on top of the refunds and the cleanup.

His takeaway, posted the same day, was disarming: “I still don’t blame Claude Code. I trusted it too much.” The failure was not the model writing bad code. It was a founder shipping payment code without a second pair of eyes on where the secrets lived. It is the single most useful paragraph in Glossa’s story for anyone building the same way: ask your AI tool whether any secret keys sit in the frontend, check it yourself, and test payments in Stripe’s test mode before strangers pay.

The breach did not kill the business. It is part of why the business is credible: a founder who posts his own security failure in public is a founder whose revenue numbers are easier to believe. The numbers below come from Stripe, not from him — but the habit of telling the truth in public is the same habit in both places.

The numbers, and where they stop

On May 6, 2026, Karbanovich posted on X that the startup had “just hit 20k+ Revenue.” By October 19, the payment data showed $14,404 in monthly recurring revenue from 100 active subscriptions, $31,173 in the last 30 days and $196,539 all-time. The figures were read from Stripe and checked on November 9 — and they stop at October 19, because the Stripe connection expired. Anything since then is unverified, and the listing says so plainly.

The split inside the numbers is the part worth reading carefully. Less than half of the last-30-days money was recurring: about $14,404 was subscriptions, and the rest of the $31,173 came from pay-as-you-go hours and one-off payments, which swing more month to month. A busy Easter looks like growth until it doesn’t. The honest way to read Glossa is as a $14.4K MRR business with a meaningful pay-as-you-go tail — not as a $31K-a-month business.

That honesty is the whole story. A real problem inside a group he belonged to, solved by hand first, built with AI tools, sold to a niche nobody else wanted, with the numbers — good and bad — posted in public. The next milestone is not a bigger launch. It is the same loop: one more church, one more Sunday, one more group that can finally follow the sermon.

The playbook

  1. Solve your own group’s weekly problem — Start where you belong. A congregation that meets every week with visitors from ten countries is a real, repeating need — a better market than a trendy one you have no connection to.
  2. Sell it by hand before you build it big — Free translation at his own church’s events came first. Two or three groups paying a small monthly fee proves demand better than a launch-day spike.
  3. Check for leaked keys before you take payments — API keys in the frontend let someone charge 175 customers $500 each. Ask your AI tool whether any secret keys sit in client-side code — then check it yourself before strangers pay.
  4. Track subscriptions and one-offs separately — Less than half of Glossa’s last-30-days money was recurring. Pay-as-you-go spikes, like a busy Easter, are not MRR — count only what repeats.
🚀 Share your milestone View Glossa →