I decided to make a small tool to help me study for my Japanese kanji exams, you can find it at https://slashie.net/kanjidrill/ The way it works is, you upload a dictionary of kanji expressions including the kanji, the furigana, and the meaning, and then it’s going to shoot random questions at you until you master them.

As I was using the app yesterday, I noticed something was missing: whenever I completed a question, I would get the correct answer, but there was a missing opportunity there to display the entire entry so that I could reinforce the learning with the remaining aspects. For instance, when missing a kanji to hiragana question, I would be shown the correct hiragana, but not the meaning of the kanji. I thought it would be useful to have that info.

At first I thought I’d do this using my morning free tier on claude.ai, but then decided… why don’t I give it a spin to do it by hand? like the good old times?

As I set to do this, and without asking, I got an autocomplete which did what I wanted. I provided no context, asked no one, and still got the job done.

It wasn’t all fairytales tho; it sought to add more styles than were needed, and in the end I had to remake a lot of it to improve the layout and make it more readable; could have been done by prompting, but at a higher cost (it was just half a cup of coffee here).

Another thing I changed was defaulting to multiple selection instead of draw/type. Again, it just feels too hardcore and it seems selecting by repetition is more effective than nailing a precise answer from memory.

Also fixed an issue with multiple choice selection for multiple valid answer entries.