Cocktails aren't described in precise terms. We reach for phrases like "bright but not too sweet", "spirit-forward with a hint of smoke", or "refreshing, but with a kick". These are subjective, multidimensional descriptions of experience - rich with meaning, but structurally fuzzy.

In this talk, I explore turning that fuzziness into something navigable. Starting with recipes and sensory descriptors, I build a feature matrix that captures multiple dimensions of a cocktail's experience: taste (sweet, sour, bitter), aroma families (citrus, herbal, smoky), texture and mouthfeel (carbonation, viscosity, astringency), alcohol strength, dilution, and even irritants such as chilli heat or menthol cooling.

The interesting work is not plotting the drinks, but deciding the axes, how to measure proxies for subjective qualities, how to scale them, and how to avoid collapsing complex experiences into simplistic categories. With that structure, we can apply dimensional reduction techniques to create interpretable maps of flavour space, and use distance-based similarity to suggest substitutions, cluster styles, and identify gaps in a cocktail "portfolio".