Level 1 Sponsor
Tastry is a multidisciplinary innovator in the fields of Artificial Intelligence and Analytical Chemistry which leverages hybridized Expert and Content-based Machine Learning and Flavor Chemistry to provide Multi-Criteria Recommender Systems.
…we like to say we “Taught a computer how to taste”.
Katerina Axelsson was born in Voronezh, Russia- immigrating to the United States when she was eight years old. She grew up and attended college in southern and central California. She completed her chemistry degree at California Polytechnic State University, San Luis Obispo, and founded The Bottlefly, Inc. (DBA “Tastry”) in 2015. Katerina is the CEO of Tastry and currently resides in San Luis Obispo, CA.
Problem. The concept of Tastry began when Katerina was finishing her degree in chemistry at Cal Poly. To pay her way through college, she worked at a ‘custom crush’ facility (a winery which makes wine for other labels). During her time there she observed many troubling realities within the wine industry: ratings, recommendations, and tasting notes are inaccurate. On one instance the facility created a large batch of wine and distributed the exact same wine to two different customers, sold by two different wineries, marketed under two different labels, and sold at two different price points: the consumer would have no way of knowing it was the same wine. Both wines went to competition where they were scored by one of the most prestigious sommeliers in the industry: one received 89 points and the other 93 points. She began to wonder. If even the world’s best sommeliers can’t give objective information about wine, how can consumers ever be expected to enter the mystical world of wine? Can wine be objectively scored? And even if it could, would everyone have the same score for each wine? The inconsistent and subjective nature of the accepted system ran contrary to the scientist in her. “There must be a more objective solution,” she thought.
Breakthrough. Katerina found support among management who allowed her to do her ‘mad scientist thing’ late in to the night for months. By analyzing hundreds of wines using various methods, she created valuable insights in the field of flavor chemistry. The next step: develop a recommendation engine. She took her results to a Computer Science PhD at Cal Poly. A half-hour meeting with one professor turned into four hours, multiple other PhDs, and an international patent attorney. The white board soon became covered in logic trees and diagrams, and they all quickly realized they were witness to a breakthrough in flavor chemistry and AI.
Solution. The initial challenge was to create a wine rating and recommendation system which was not only objective but would also take into account consumers’ preferences and aversions as well thereby creating a rating system personalized to the individual or group of individuals.
Example: Tastry might recommend a Pinot Noir to John which is a 94% match, but which may only be a 76% match to his wife Joan, and an 81% match to their daughter Jennifer. On the other hand, John can use Tastry to search for the best wine for all three of them- perhaps finding a Syrah which yields match scores of 89%, 92% and 90% respectively; i.e. the best choice for the group.
The efficacy of the Tastry system has been substantiated both in-house and by independent parties, and the Tastry methods are proving directly applicable and viable in many other sensory-based products from perfume to cannabis to food.
Sensory-based. Recommender Systems typically employ some form or combination of demographic, collaborative, or generalized content filtering, but these types of algorithms don’t work for sensory-based products. Tastry uses analytical chemistry and automated feature engineering to define the palate profile of the consumer and match them to the flavor profile of a product.
We know if a consumer will like the taste of a wine, or the scent of a perfume, before they try it – and Tastry does not require a consumer to ‘rate’ any portion of an existing inventory to derive insight.
Flavor Matrix. Tastry has innovated novel methods to chemically analyze and mathematically describe the Flavor Matrix. There may be few (or hundreds) of compounds in a sensory-based product. Understanding how each of those compounds in various concentrations impacts every other compound (by masking or expressing various notes or flavors) defines the Flavor Matrix.
Example: Virtually every red wine has the compounds responsible for the taste of cherry, but not every red wine expresses the taste of cherry.
Differentiated Sensory Profile. Tastry developed the Differentiated Sensory Profile “DSP”: a mathematical descriptor which expresses the palate profile of a human consumer as it relates to the Flavor Matrix. The nature of the DSP allows Tastry to extrapolate palate preferences across multiple product types mitigating the Cold Start challenges typical of Knowledge-based systems.
Example: If Tastry understands a consumer’s wine palate, we can predict that consumer’s palate for other products with a high degree of accuracy; any subsequent additional product specific palate data will augment the AI but are not necessary to provide meaningful initial predictions.
Current. The current Tastry business model is primarily B2B2C. Consumers create a Tastry account, answer a of number sensory preference questions, and the recommender engine matches the consumer to available products at a brick and mortar retail location or online inventory. In addition to wine recommendations, Tastry provides cheese, cracker, meat, and recipe pairings (filterable by various aspects such as price). The current model provides the retailer a positive ROI “Return on Investment”, increased margin, and gross revenue all while providing increased customer engagement, satisfaction, and loyalty. Tastry is a win-win for the retailer and the consumer.
Future. Tastry has increasingly been pursuing a B2B model for improving the design and development process of other businesses. Tastry intelligence can be used by manufacturers to not only standardize and codify the ‘sensory language’ but also in the development of sensory-based products through market segmentation of a population or demographic by flavor preference. This will further expand on the creation of ‘crowd-pleasers’ or creating ‘one-off’ consumer specific recommendations and products which considers individually relevant health and wellness attributes of consumable sensory-based products.
Big Picture. Additionally, our AI innovations in system architecture and methods are often described as revolutionary and able to increase the efficacy of many forms of recommender systems.