Put this cup of coffee into words
Every taste has its own vocabulary. flavorwords is a free coffee-tasting web app.
Most people can tell that two coffees taste different but not say how. flavorwords asks a few plain questions about the cup in front of you and helps you find flavor words that fit — grapefruit, jasmine, dark chocolate, cypress — then puts them on a flavor card.
How it works
- Say how the cup was made. Brewing method, roast, variety (single or a blend of up to three), process and, if known, origin. This gives a starting point from the review corpus, never a verdict.
- Answer up to six plain questions. Acidity and fruit, the main aroma, sweetness, mouthfeel, then at most two follow-ups chosen by your answers (which citrus or berry exactly, bitterness, aftertaste, how clear the aroma is, the overall impression). Four options a question, five for the first two; “I can’t say” is a regular option and changes nothing. The options differ by kind of coffee and none refers back to an earlier question.
- Keep three to five of eight concrete flavor words. The card follows you first: the word you chose leads, then the kind you pointed to; the data speaks only where you said nothing.
- One more step when your answers conflict. Whatever you tick there appears on the card, and the words you picked yourself are kept.
- Get a flavor card. The context, the words you confirmed, evaluation rows (body, roast and bitterness, aroma, aftertaste), the nearest of sixteen reference profiles, and where each sentence comes from. Export it as an image. Any question can be reopened from the progress bar; the other answers are kept and recomputed.
English and Chinese share one data backend. It works offline and installs to the home screen on iOS, Android and HarmonyOS. No account, no server-side state, nothing you answer leaves your device, and no generative model runs.
Where the data comes from
9,128 coffee records and 83,031 flavor descriptions (77,637 after cleaning) from eight professional review sources — CoffeeReview editorial reviews, Cup of Excellence juries, two Q-grader datasets and four smaller panels — normalised to 94 sensory concepts on 12 flavor dimensions (acidity, sweetness, body, floral, fruity, nutty and chocolate, fermented and winey, roast and bitter, spice, herbal and green, woody and earthy, defect) and clustered offline into 16 reference profiles; the app carries 101 flavor words in English and Chinese. No source text is stored and no commercial coffee is named. At run time the app does deterministic vector matching in the browser; every output is traceable to a row or a rule.
Who made it
Research, design and development: Dai Pan (潘岱), June–September 2026, individually led with AI-assisted implementation. Expert interviews, a cupping study, two questionnaire rounds and two pre-launch tasting tests shaped the product; it has not been validated with a general-audience sample.
Source, data guide and method on GitHub · plain-text summary for language models