Vibe coding

It looks like I’m back into making apps. After making over 100 apps with AI, I took a long break. But today I put together a new app with AI–what people are now calling vibe coding.

I’m building on my project to create scenes for reading Japanese. The app lets the user pick out which parts they want to be written in Japanese and in what format to write them. With Japanese you can write the text in a few different ways.

• Roman letters (romaji),
• use the hiragana script
• a mix of both hiragana and katakana (kana)
• in the more complex Chinese characters (kanji)
• kanji with hiragana written above the characters (furigana)

{
"id": "s11",
"role": "dialogue",
"english": "\"I bet there's a secret door.\"",
"japanese_kanji": "「たぶん、秘密の扉があると思う。」",
"japanese_kana": "「たぶん、ひみつの とびらが あると おもう。」",
"japanese_furigana": "「たぶん、<ruby>秘密<rt>ひみつ</rt></ruby>の<ruby>扉<rt>とびら</rt></ruby>が あると おもう。」",
"japanese_hiragana": "「たぶん、ひみつの とびらが あると おもう。」",
"japanese_romaji": "「Tabun, himitsu no tobira ga aru to omou.」"
},

ChatGPT helped me plan the app and set up the data format. The scenes were all written by ChatGPT, but I used Gemini for for the translation and formatting as JSON because it used the context window a bit better for this task. The images are generated by Ideogram. The code was created by WebSim using Claude 3.7, and edits were made with Cursor using Claude 3.7. Whew!

Now that it’s all set up I can just add more data to populate more scenes. But that’s a project for another day.

• Japanese Scene Reader: A tool for reading Japanese scenes at your level of proficiency.