A cocktail laboratory across web and native mobile.
You taste something memorable at a bar, take a rough guess at the ingredients, and then start iterating at home - but most tools stop at storing recipes instead of helping you learn toward a better version.
Dram combines AI menu scanning, attempt logging, recipe history, cabinet tracking, and triangulation into one workflow so every experiment improves the next one.
The project now spans both web and an Expo-based iOS and Android client, carrying the same core product idea into a form factor that works at the bar, in the kitchen, or while shopping for ingredients.
Instead of saving a drink as a vague memory, users can turn a menu photo or URL into a structured recipe guess with ingredient roles and confidence cues.
The first draft of the drink is generated, not guessed alone.
Tasting notes, substitutions, ratings, and versions all compound so the best-known recipe gets sharper as the user experiments across time.
The system improves because the cook keeps trying.
Ingredient tracking, availability awareness, and batch scaling keep experimentation tied to the practical reality of what the user can actually make tonight.
More laboratory than scrapbook.
Claude-powered image and URL parsing turns bar menus into structured cocktail records, with ingredient roles, confidence levels, and recipe estimates ready for iteration.
Every strong attempt improves the system estimate, weighting higher-rated versions more heavily so the best-guess recipe becomes more useful over time.
Recipe history, flavor architecture, cabinet management, and batch scaling turn the product from a static recipe saver into an actual experimentation environment.
Dram is built around the cycle of noticing a drink, attempting it, and then converging on something better than the original memory.
Menus, photos, and links are translated into structured recipe candidates so the user starts with something concrete instead of vague recall.
Each version records adjustments, impressions, and outcomes so experimentation becomes cumulative rather than disposable.
Higher-quality attempts pull the estimate closer to a useful recipe while cabinet data and scaling keep the result actionable.