Flavor Genome Project
Project focus: How Might We Categorize the Sensory, Emotional, and Familiar (Cultural) Experiences of Flavors?
Recognition: Mostly grunt work, little glory, but exciting what will be possible in the future. This project later powered some of the scoring mechanisms of the Chef League iOS App.
2015 to present
This project is spearheaded by Dr. Altringer of Harvard University’s School of Engineering & Applied Sciences. The Flavor Genome Project is part of a longitudinal project that explores how components of flavor combine to create delightful multi-sensory, chemical, emotional, and cultural experiences.
Future – We are currently seeking collaborations, partnerships, and funding for the next stages of our growth. Contact: ba[at]seas.harvard.edu.
Consulting – FGP has grown into a very large resource. Consulting services help fund and grow the project. As of 2019, we now process millions of data points that connect multiple databases: large scale recipe data, natural language recipe review data, quantitative recipe review data, ingredient data (including flavor compounds, nutrition information, and dozens of flavor and texture attributes), historical chef preferences data, and ingredient adjacency data for global recipes.
*WHAT WE DO: COMPUTATIONAL FLAVOR DISCOVERY*
The Flavor Genome Project is automating our understanding of what people are intuitively searching for in a food or drink experience, regardless of the language they use to search for it, intelligently understanding flavor goals in context, and, ultimately, making it easier for people to discover experiences they are likely to enjoy.
We are developing ways to computationally understand everything about how people experience flavor (how everyday people describe flavor experiences, how experts describe the same flavor experiences, the sensory experience of what it tastes like) by growing and analyzing very large datasets. We match this information to data on flavor compounds and nutritional information contained in the recipes under review.
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. An issue with current recipe recommendation systems is that, although there are algorithmic ways of suggesting that two things taste alike, this does not mean that they will taste alike to different groups of real people.
*RELATED NEWS AND PARTNER NEWS*
FGP Founder becomes launch partner for Off Their Plate – Offtheirplate.org is a COVID-19 relief initiative for restaurant workers and emergency workers. The team raised $6.5 million in four months to restore restaurant workers’ wages by preparing meals for emergency workers. Check out press for Off Their Plate in the New York Times, Boston Globe, NBC, Seattle Times, and more.
FGP is pairing wines with new dishes by 25 top chefs for Boston-based delivery services – We partnered with a pop-up concierge food and beverage delivery service that is bringing top chefs’ dishes with beverage pairings to your door.
FGP promotes Project Restore Us – COVID-19 has been extremely difficult on everyone. Boston-based Project Restore Us is chef-led and designed to aid those who are last in line — working immigrant families with significant barriers to food access. They are teaming up with restaurants, who know firsthand how to procure food safely and how to feed people efficiently. The project seeks help or donations if you are able.
FGP Chef League launched using Flavor Genome Project scoring algorithms – It’s the first time it has been possible to compete to make recipes taste better in a mobile game. Chef League is the first public product based on the Flavor Genome Project data. Check out Chef League, the first iOS mobile game to let players compete on making things taste better. Most cooking games on the App Store are actually time management games (e.g., serve a burger to one person, now two, now 15 people). Chef League, powered by the Flavor Genome Project databases, is the first time we can score millions of flavor combinations in real time. Players compete for fun to fix recipes and master the functional use of salt, sweetness, acidity, fat, spice, and more. Show off your creativity, discover new ingredients, and learn from chefs. The game was designed in collaboration with professional chefs and flavor scientists, as well as real data and reviews of over a million recipes.
Press for Chef League:
Venturebeat: “Chef League challenges you to improvise like the pros”
Harvard SEAS: “Cooking up an educational competition”
Tapsmart: “Solving flavor problems for fun”
Hackernoon: “A new app is transforming learning through AI”
SXSW Chef League Launches for Pre-Order on iOS, March 2019 – The first product based on the Flavor Genome Project data launched for pre-order. Check out Chef League, an iOS mobile game where you compete for fun to fix recipes and master the functional use of salt, sweetness, acidity, fat, spice, and more. Show off your creativity, discover new ingredients, and learn from chefs. The game is designed with flavor scientists, based on real data from hundreds of chefs and over a million reviews of thousands of recipes.
Flavor Fluency Film debuts at MIT Cambridge Science Festival, sponsored by Harvard University, Early 2018 – In April (details here) as part of the Cambridge Science Festival, Altringer and Chef Tracy Chang hosted an afternoon exploring how five Boston chefs ‘design’ flavor experiences. The event featured the debut of a short flavor film Flavor Fluency featuring chef-owners Chang (Pagu), Michael Pagliarini (Giulia, Benedetto), Tse Wei Lim, Diana Kudayarova, and Sam Treadway (Backbar). After the viewing, guests tried their own hand at thinking like these chefs by experimenting with acids, umami, fats, sweetness and bitterness with our special snack and beverage pairings (food, wine, beer).
MIT Museum, Late 2017 – collaborated on AI and Flavor with the MIT Museum for the sold-out Food for Thought exhibition. Chef Tracy Chang developed four menu items and Altringer’s software generated 2-3 beverage pairings to each item. Pairings deviated substantially from human-generated pairing stereotypes, which tend to assume things like ‘if it grows together, it goes together’. Software-suggested pairings like Porter and Stout with a chocolate ganache dessert with peppercorns surprised and delighted guests. The blind parallel tastings showcased different ways that wine, beer, and cocktails can pair to the same food dish, with interactive effects that highlight different flavors in the food.
Software versus Sommeliers, Mid 2017 – tested a later version of the software, co-hosting a blind pairing dinner with Chef Tracy Chang of Pagu. Several sommeliers provided pairing suggestions to Tracy’s seven-course menu and guests blind-tasted the sommelier vs. software beverage pairings.
On two of the seven dishes, the software and sommelier suggestions were very similar (providing a control group) and indeed guests could not tell the difference. For two additional pairings, the crowd was split, with half preferring the slightly sweeter or slightly more acidic suggestions. On a fifth dish, the software interpreted charcuterie and cheese as mainly meat, and suggested a wine accordingly. The sommeliers interpreted the charcuterie and cheese platter as mainly cheese. The crowd split, depending on what they preferred eating. On the remaining two pairings, the guests preferred the software-suggested wine. The final scores: control courses (2 pairings): tie; split preferences courses (3 pairings): tie, strong preference courses (2 pairings): software wins both.
‘Feelings for Dinner’ chef and sommelier collaboration, Mid 2015 – collaborated with chefs Tse Wei Lim and Diana Kudajarova to test the earliest version of software developed from FGP. The collaboration helped show that the approach could inspire novel professional culinary designs. It led to ‘Delightfully Paired: Feelings for Dinner’, an event in Cambridge, MA in which guests explored – through taste – whether food and wine pairing can tell an emotional story.
Database Development Begins, Early 2015 – began building an interactive catalogue of 200,000 multi-sensory pairings in daily life (foods, beverages, emotions and more). By 2019, the Flavor Genome Project would be processing over a billion data points.
Sensory Design Residency, Mid to late 2014 – Through a design residency in Italy, Altringer explored approaches to deconstructing flavor experiences with the Sensory Composition project.
The initial idea for the project has roots in Altringer’s experience training for international tasting competitions with the Cambridge University Blind Wine Tasting Society from 2007-2010. In 2008, Altringer was a top four finalist for the Lafite-Rothschild competition, and went on to win the 2009 Bollinger competition. The initial inspiration for the Flavor Genome Project was to extend human memory because professional tasting training helps people intuitively experiment with new flavors, but it is difficult to remember all possible flavor data.