Paste a roaster's product page. The ledger reads it, predicts quality and a fair price, and judges the asking price.
JEV extraction, normalization & prediction
TypeSafe JEV answers typed questions about origin, process, variety, and the coffee's story. The app copies listed prices and bag sizes from product data, with source snippets for review. Pages without one coffee product are declined.
Currencies convert to USD at live rates and bag sizes to grams. When the page omits a roaster country, the app uses domain evidence. The appraisal record lists its assumptions.
A lightweight local model estimates quality. The fair-price model uses the same JEV feature schema in training and serving, then compares its estimate with the listed price.