How Non-Developers Can Take a Niche Learning App All the Way to Launch

It worked in two days, but finishing the screens took three weeks

Mikko Ikola built Mandarin Melody AI, an iPhone app that creates personalised songs from Chinese vocabulary. When users enter words from a Chinese proficiency test or words they choose themselves, the app produces songs and karaoke-style lyrics showing Chinese characters, pinyin, and English meanings.

He described the idea to a conversational AI tool, then put the result into Figma Make, a screen-building tool. The tool created onboarding, word selection, music generation, and playback screens. A first version running on an iPhone was ready two days later.

But the work broke down when it was time to turn the screens into a real app. A method intended to connect the screen-building tool directly to the programming tool did not work properly, so he had to capture each screen and upload them one by one again to Codex, which creates program files.

After the first version, he still needed to set up a place outside the app to store and process data, connect a music-generation service, and fix bugs on real iPhones. The working draft took two days. A version ready for release took about two weeks, or about three weeks including final screen work.

The maker’s conversational AI subscription started at about KRW 27,000 per month and rose to about KRW 134,000 during the more complex music-integration work. Apple Developer Program enrolment for App Store distribution costs about KRW 133,000 per year. The actual total cost of the screen tool, data storage, and music generation was not disclosed. The stated figure of about KRW 200 per song was an assumption used to explain commercial viability.

What if course materials only had to be entered once?

Follow one hypothetical user. She teaches an evening Chinese proficiency test level 3 class with 18 students. Each week, she sends self-made vocabulary materials and audio files through KakaoTalk (카카오톡), a Korean messaging service.

She first selects 30 words for the week in Excel. She pastes the list into a conversational AI tool to create example sentences, moves it again to a music-generation site to make a song, then separately uploads lyrics and audio files to documents and KakaoTalk.

The same vocabulary list passes through at least four places. If one character is wrong or the word order changes, she must check again which file is current. To reflect words students got wrong in the following week’s materials, she must manually collect the results again.

A service that connects this workflow could let the instructor upload an Excel file once on the first screen, then choose only the class name, test level, and learning goal for the week. The service could create vocabulary cards, example sentences, song lyrics, an audio draft, and short check questions as one set from the same word list.

When the instructor edits one line of lyrics, that edit could also appear in the vocabulary cards and questions. Only approved materials would go to the student app, and words with many incorrect answers could return to the following week’s review list.

Launch preparation could continue in the same place. The operator could fill in App Store screenshots, description text, the list of personal information collected, whether payments are used, and a test account for review, then receive a submission draft and a list of missing items.

Still, much remains in human hands. The instructor must decide whether example sentences fit the class level, whether Chinese tones and pronunciation are correct, and whether music and image usage rights are acceptable. The app operator is also responsible for privacy language, payment methods, blocking inappropriate generated material, and replies to Apple’s review team.

Services that connect narrow learning behaviours

Speak, a US company, launched first in Korea in 2019 and focused on one problem: helping people who can read English actually speak it. It provides pronunciation and expression feedback on users’ voice conversations, along with personalised lessons, and charges a subscription fee.

In a 2024 announcement, the company said it had more than 10 million users in more than 40 countries. This is a company claim, but it is an example of reducing the break between speaking practice and feedback, rather than simply accumulating more learning materials, across multiple countries.

Drops, which started in Estonia, focused on five-minute daily vocabulary learning instead of long lessons. It uses images, associations, and swipe games for short vocabulary sessions, and sells subscriptions for access to more learning.

When Kahoot! acquired Drops in 2020, the service had 42 languages and 25 million cumulative users. In the first quarter of 2021, about 130,000 paid subscriptions were counted, showing that a narrow starting point of short learning time and less commonly served languages can also become a paid service.

MySwimPro, from the United States, chose one sport rather than language learning: swimming. It provides personalised training plans and an exercise library, and charges a subscription fee.

When Apple featured the service in 2020, it was serving about 50,000 active users with a team of 12. Rather than building a general fitness app, it focused on telling swimmers what to do next in the pool.

Things you could build now

1. A course-sheet-to-learning-app builder

• What it does: Turns one Excel vocabulary sheet into cards, example sentences, audio, questions, and a review schedule, then sends them to a learner app.
• Who uses it: An academy instructor running one or two Chinese proficiency test classes and sending self-made materials through KakaoTalk each week.
• Why now: Screen and program drafts can be made quickly, but people still re-enter the same words into several tools.
• First screen: Show only file upload, class name, test level, and the number of words for the week.

2. A launch-preparation assistant for small education apps

• What it does: Takes a working app and prepares its app marketplace description, screenshots, privacy items, review account, and pre-submission checklist in one place.
• Who uses it: A non-developer instructor who runs certification courses and has made a first version of a practice-question app.
• Why now: App-building time has fallen, but Apple still reviews simple web wrappers, unfinished features, and inaccurate privacy disclosures.
• First screen: Ask four questions: whether login is required, whether payments are used, what users upload, and what information is sent outside the app.

3. An operating dashboard that prevents generation costs from overrunning revenue

• What it does: Records the cost of every generated song or voice clip, and warns when free usage and paid pricing could create losses.
• Who uses it: A solo operator of a small learning app offering pronunciation songs or personalised audio.
• Why now: The cost of music and voice generated by users each day can grow faster than the cost of building the app.
• First screen: Include fields for monthly fee, expected user count, generations per person, and cost per generation.

4. A checking service that compares design screens with the real app

• What it does: Places the original design screen beside the real iPhone screen and creates a fix list for differences in text, spacing, and button behaviour.
• Who uses it: A solo product planner who created screens in a design tool and handed app implementation to an external maker or AI tool.
• Why now: In the Mandarin Melody build, a direct connection between tools was blocked, so screens had to be captured and sent again.
• First screen: Show two large upload areas, one for the design screen and one for the real app screen.

What to check today

Ask one Chinese-language instructor for a 30-minute screen-sharing session. Count every place where the next lesson’s vocabulary sheet is copied before it reaches students. If the same list moves through three or more places and this happens every week, it may be worth testing a service that removes that movement before adding new learning features.

Why this matters where you are

AI tools may shorten the time needed to create a working app draft, but they do not remove the handoffs between teaching materials, generation tools, app implementation, and marketplace review. Check where one source list is copied across tools in your own market, and which review requirements remain the operator’s responsibility. A narrow workflow that removes repeated transfer may be more useful to test than a broader learning product.