Fragrance has always had a discovery problem. You cannot smell a perfume online, the language of scent is vague, and catalogues run into thousands of options. Artificial intelligence does not give screens a nose – but it is changing how people describe, search for and find fragrance.
From keywords to conversation
Traditional search needs exact words. Most people, though, describe scent the way they feel it: “something like a hotel lobby”, “warm but not sweet”, “what my grandfather wore”. Language models are good at turning these fuzzy descriptions into structured attributes – notes, families, moods, seasons – that a catalogue can actually be searched by.
Matching by profile, not by name
Once fragrances are described as data – accords, top, heart and base notes, intensity – AI can compare them mathematically. That makes it possible to say, “you love this perfume; here are the oils with the closest profile”, even when names are completely different.
Seeing a bottle, not just reading a name
Computer vision and text recognition now let a phone read a perfume label from a photo. It is not perfect – stylised labels, reflections and small print are hard – but combined with a large reference database it can identify many bottles in seconds.
What AI does well – and what it doesn’t
- Does well: narrowing thousands of options to a shortlist, understanding natural descriptions, remembering preferences, answering questions at any hour.
- Doesn’t replace: the nose. Final judgement on how an oil performs on skin or in a candle still belongs to people and perfumers.
Where Haveli AI fits in
We built Haveli AI to take the friction out of finding the right oil. Describe a note, a mood, a budget or a perfume you love – by text or voice – and it ranks the closest matches from our 6000+ oil library. Our Perfume Decoder reads a bottle from a photo and learns from confirmed answers. Then our perfumers take over: every match is backed by oils compounded and checked on the bench in Mumbai.
Try it now: ask Haveli AI.