I built Riddle of the day, a bilingual daily logic puzzle site:
https://acertijodeldia.com/en/
The product is intentionally simple: one carefully chosen puzzle a day, with hints, explanations, streaks, and spoiler-free challenge links. No account, no ads, no feed.
The technically interesting part is answer checking.
For most puzzle sites, the user either selects from multiple choice or enters an exact expected answer. I wanted something more natural: users can write an open-ended answer, and the system tries to classify it as correct, partial, incomplete, or worth refining.
Today’s puzzle is a good example:
https://acertijodeldia.com/en/riddles/the-twenty-five-horses/
The challenge is not just getting “the number”; it is explaining why the strategy works and why fewer races cannot guarantee the result.
I’m still refining the balance between:
- accepting equivalent explanations;
- not giving false positives;
- giving useful feedback without spoiling the solution;
- keeping the experience calm and fast.
I’d be interested in feedback from people who have built similar natural-language evaluation flows.