Our summary
Pochang Lab built a Japanese word game to experience Jev’s speed and cost, rather than just read a price list. Players enter technology terms to fill five category meters. Jev answers set questions without writing replies, while the screen shows the requests, response times and estimated running cost.
Five questions about each term travel together in one request. Jev judges its category, characteristics and which prepared trivia entry fits. Ordinary code extracts words, keeps score, checks repeats and selects the displayed lines. Failed questions about repeated text and related technology names helped the maker redraw that boundary.
The article’s demo reports a 407-millisecond average and about 0.15 yen in usage. Its 14 judgments include two local checks and twelve calls to Jev. These are maker measurements, not guarantees. The cost uses a fixed exchange rate; typed text goes to TypeSafe and is also logged locally.
Key takeaways
- Use Jev for fuzzy judgments, but keep exact repeat checks, point calculations and prepared replies in code.
- Send independent questions about the same term together, and distinguish naming a technology from naming something related to it.
- Count actual service calls separately from local checks; reported speed and estimated usage cost do not establish accuracy or total development cost.