Show HN: Context-aware Japanese furigana using Sudachi and ModernBERT

A new Hacker News project promising to add little pronunciation guides above tricky Japanese characters has landed with a very online mix of adoration, nitpicking, and instant stress-testing. The pitch is simple and very appealing: paste in Japanese text, or upload files like PDFs and ebooks, and the tool figures out how hard words should be read — all for free, no account needed, and with uploads deleted within 24 hours. For language learners, that last part got almost as much love as the feature itself. One commenter basically gave it the holy blessing of the internet: "It really works. Very cool." Another praised the no-signup setup like it was a rare mythical beast.

But of course, the real action was in the comments, where admiration quickly turned into a live-fire exam. A Japanese commenter said they were genuinely impressed it handled notoriously difficult place names, then immediately brought receipts for a reading that seemed off: the kind of feedback that says, "great job, now let’s see if you survive the boss battle." Another user found an even messier stumble, reporting that a phrase came out as something wildly wrong — the sort of bug report that doubles as a jump scare.

The vibe? People want this badly, especially learners frustrated with clunky older tools, but they also want it to be really right. There was even a wish for pitch-accent support, because on the internet, every useful tool is only five minutes away from becoming a debate about the next impossible feature.