Creating JaptoAnki

I loved RTK, and finished it within a couple of months of starting it. And as I was practicing with the community made anki decks for daily practice I became increasingly frustrated that the Heisig stories weren’t displayed in the cards. Eventually I found this deck: in which has hocanhlinks for each kanji (these links, have the koohii community stories made for going through the RTK book). And that was amazing, but I really disliked having to click on the links manually each time and all this micro-friction in just trying to memorise and practice some Kanji. Also, all of these decks were missing a crucial point which was the actual reading of the kanji. Sometimes the kun-yomi and on-yomi would be there (but what use is there in having those if there’s no word there?) This proved to be quite a challenge actually because the script now needs jaconv (or an alternative), fugashi, tqdm, etc to provide the actual reading. And the reading would have to be displayed as furigana (which is usually easy to display in Anki, but when the html for the cards starts to get complicated was a enormous pain in the ass)

Anyway, I eventually vibe-coded the beginnings of JaptoAnki which was this simple python script which would use Manga-OCR (cause I didn’t know about Mocuro until wayyy later unfortunately) to extract kanji compound words from manga. This script sucked so unbelievably bad, because mangaOCR is only meant to recognise manga text in a bounding box. As in, not the entire fucking page all together. I ignored this obvious CRUCIAL issue thinking that it was a issue with my code. So I made the script connect to Anki through ankiconnect so that the extracted kanji would go straight into my anki deck. The cards themselves I defined? created the models? in anki with hocanhlinks (basically copying the other deck from earlier) for each kanji within the word / sentence.

Of course, I never actually used the script and just kept trying to make it work (STILL USING MANGA-OCR ON THE ENTIRE PAGE BTW), adding features like making new decks and adding tags through flags, a user .txt file in ~/ where users could list their “mastered kanji” so that anything extracted from manga which contained only kanji within “mastered kanji” would be ignored, and “mastered kanji” would never produce hocanhlinks.

It was just about now that I finally looked into Manga-OCR and found the obvious major issue. So I looked about, and found Mokuro-OCR and was devastated to find that the people at Mokuro made pretty much exactly what I was looking for in an OCR. After switching to Mokuro and adding google translate capability (which, to avoid paying, the script actually bundles 10 sentences / words together and sends to google-translate for translation at a time, before waiting and repeating with the next batch) the script started becoming an actual usable tool.

I changed from a cli tool to something more UI like with questionary, allowing for file navigation, and choosing of files that way. The issue with navigating with questionary is that any directory with < 36? options can’t use the keyboard shortcuts meaning you have to use the arrow keys painfully slowly. Questionary remained for quite a while before I asked an AI to vibecode me a yazi-style navigation which it did perfectly first try… fuck