Flavor Genome Project

Since 2015, the Flavor Genome Project has been using mixed methods to explore how components of flavor combine to create delightful multi-sensory, chemical, emotional, and cultural experiences. To date the project has processed more than 4 million recipes, together with the reviews people wrote about them, the flavor compounds and nutrition of their ingredients, and the preferences of chefs across history.

We use natural language processing to identify new data-driven patterns or ‘genres’ of flavor experiences (beyond categorizing food as ‘Italian’, ‘Mexican’, etc) based on how ingredients and dishes are described by experts, scientists, and reviewers.

The initial idea has roots in training for international tasting competitions with the Cambridge University Blind Wine Tasting Society (2007–2010), where I was a top-four finalist for the Lafite-Rothschild competition and won the 2009 Bollinger competition. The inspiration was to extend human memory — professional tasting training helps people intuitively experiment with new flavors, but it’s difficult to remember all possible flavor data.

In 2017, the software was blind-tested against sommeliers in a pairing dinner with Chef Tracy Chang of Pagu. The software won both strong-preference courses. Later that year, a collaboration with the MIT Museum for the sold-out Food for Thought exhibition showcased software-generated pairings that ‘surprised and delighted guests.’

Early forms of the project include InJoy (2015) and the Amalgam Flavor Pairing API (2016). Feelings for Dinner (2015), a collaboration with chefs Tse Wei Lim and Diana Kudajarova, tested the earliest version of the pairing software.