Bukvyk

My son was learning to read, so I started building a reading app for him. About six months before this build I tried making one in Figma Make, but the tools weren’t good enough yet. This autumn I came back to it with the content I’d already gathered. Bukvyk teaches children aged 4–6 to read, sounds first: 14 short games from sounds to syllables, words and sentences, plus short readings and bedtime tales. It’s for families in Ukraine and for the 5.7–5.9 million Ukrainians abroad, whose children learn to read a language they don’t hear at school.

From a sound to a syllable, a word, a sentence and a tale.

The research made sound a release blocker

On day two the app looked finished, with eight games and bright emoji, but it was silent. The research gave an answer I didn’t like: the two strongest predictors of reading are both about hearing (National Early Literacy Panel, ~300 studies), so a silent app trains neither. Sound went from a backlog item to a release blocker: 894 clips, each letter’s sound separate from its name (a child needs [m], not “em”), bundled so the app works offline.

Every word, syllable and letter sound is voiced and bundled, so the app works offline.

What I cut

• 42 illustrated scenes. “True or false” needs a sentence that’s reliably false, and a still picture can’t give one. The game was rebuilt on emoji.
• Narration on the short readings. A child who has heard a text retells it instead of reading, so the progress chart would show growth where there was none.
• Four finished games that could be solved without listening.

Some of the 42 generated scenes that were cut.

The emoji versions that replaced them.

Designed for someone who can’t read yet

• Emoji instead of illustrations. I tried photos first. Emoji turned out more childlike and easier to understand. I went through every open-source set and chose Microsoft Fluent: 3D, wide coverage, animated, and one consistent look across about 500 pictures, which I couldn’t have drawn myself. Type is Onest, a plain grotesque without serifs.
• No red for a wrong answer. The picture shakes and a sound plays. Green is reserved for “right”.
• No second right answer. 🐈 and 🐱 pass a duplicate check, but a child sees two right answers to “cat”, so look-alikes are grouped by hand.
• Nothing is locked. Siblings share one tablet and a child should be able to peek ahead, so levels are home-screen steps and each game adapts to the child.

The app’s stickers, with the same holographic shader as in the app: Б-words that keep changing. Move the pointer over them.

A free version for classrooms

The first teachers who replied had a whiteboard and a laptop, not twenty phones. The plan: teachers use it in class and recommend the paid app to parents. Because the app is Flutter, the browser version took a day: one link per school and none of the home app’s purchases, progress or stickers, so it doesn’t replace buying it. In a class, auto-difficulty would average every child, so the teacher gets a one-tap level button. I interviewed five teachers: showed them the app and asked what was wrong. Their notes mixed learning with taste (louder music, different pictures). I took only what mattered for learning and the school programme. That gave version 1.1.2: the alphabet in primer order, a separate letter order for English-speaking children, many words removed or replaced, and closed syllables taught as the reverse of open ones (МА, then АМ). A first-grade class in Kyiv now uses it on a whiteboard.

The browser version for a classroom whiteboard.

Version 1.1.2 after the teacher interviews: letters in primer order, open and closed syllables.

One price, no subscription

One purchase, with no backend, accounts or analytics. Ukrainian competitors charge $1.99–2.99 and full versions sell at $4.99, so in Ukraine, where people can pay less, it’s $4.99. Abroad, where a paying family is worth $25–32 a year, it’s $9.99.

There’s no subscription because a child learns to read once, and a parent would cancel as soon as that happens.

Progress for the parent, and one purchase for the full version. (TestFlight’s sandbox shows the price in lira.)

How I worked

I own the research, scope, cuts, copy and the final call. AI agents in Claude Code write the code. The tales were written by agents to rules I set, the cut scenes were AI-generated, and the voice is synthesised (ElevenLabs), with takes I pick by ear.

Where it is now

Live on the App Store since October 17, 2026 and on Google Play since October 21, after two Apple rejections under its Kids Category rules. I sent 6,338 letters with a classroom code. By November 5, the link had been opened from 504 of them (8%), and 138 came back on another day. That 504 doesn’t count links that antivirus software opened by itself as soon as the letter arrived: on one domain it “opened” 87% of them within hours. At first I wrote those letters with an LLM and proofread them. An important contact told me it showed. Now I write personal letters myself.

The home app is still small: 106 downloads on iOS by November 2. On Google Play, 955 page views in 28 days turned into 67 installs, and 61 of the 67 devices with the app on November 2 were in Ukraine. Four people bought the full version, three on iOS and one on Android: about 2% of the 173 installs. A father in the Netherlands wrote that his seven-year-old daughter, who has lived abroad all her conscious life, now does the exercises with interest and writes in Ukrainian with more enthusiasm.

The store listing.

What I’d do differently

• Research before the prototype. I built eight games before the research showed they needed sound.
• Record the sounds with a person. My own research said synthetic voices are weak on isolated sounds.
• Interview teachers and parents before building. The five teacher interviews came after launch, and I still haven’t interviewed parents.