AromaVerse

The Fragrance Shop runs the world's first algorithmic perfumery machine. You fill in a questionnaire and a machine mixes three 5 ml perfumes for you. It works, but you can't smell anything until the bottles are made. The experience is a form on a screen. Nothing to smell, nothing to watch, and only one person can play at once. For a shop that sells scent, the process felt kinda scentless.

AromaVerse makes scent something you can hold. Eight essential scents became eight modules, one shape and one colour each: ylang-ylang, mandarin, grapefruit, lemon, lavandin, eucalyptus, peppermint, and bergamot.

The shapes came from people, not from us. We sketched forms for every scent and had participants vote by smell. Same result as the bouba and kiki test: round forms read as soft and calming, spiky ones as sharp and fresh. Three generations of shapes later, we 3D printed masters, made silicone moulds, and cast each module by hand.

Then the interactive part. A camera watches the table. An object-detection model recognises which modules you put down, and TouchDesigner turns your blend into live visuals on a big screen. Pick shapes together, watch the mix react. We tested two earlier concepts on users first, and the lesson was clear: people don't care about the tech, they want to take part. So no questionnaire. Just shapes and scents on a table.

A team project with Ya Wang, Jiaxin Wang, and Tiancong Wang.

The demo worked. Detection ran at 88 to 98 percent confidence across the modules, live, while people moved them around. We showed the project to Frederik of Scentronix, who builds the algorithmic perfumery machines. He liked the tangible shapes and wants the visual language to match most people's intuition. The next step would be piloting it in a real store.

Being honest, the shape-to-scent mapping is only half proven. We tested it on a small group, and scent is personal. And the demo still can't make you smell anything. That part stays analogue.