Halla, an artificial intelligence company used to personalize online grocery shopping, received a $1.4 million seed fund this spring from a few notable investors, including SOSV and E&A Venture Capital.
The startup, co-founded by UC Berkeley alumnus Henry Michaelson and his colleagues from USC and UCLA, aggregates data to personalize online grocery shopping, an industry that is expected to grow dramatically in the coming years.
“We work with grocers, both online and in stores, to take this mess of data and to be able to understand all the hidden correlations … between products,” said Michaelson, who serves as the chief technology officer and president of Halla.
According to Gabriel Nipote, co-founder and chief operating officer of Halla, U.S. online grocery sales are expected to grow into a $100 billion industry by the year 2022, up from the industry’s 2017 value of $14 billion.
Halla was borne out of a collaboration between childhood friends Nipote, Michaelson and Halla CEO Spencer Price. The three knew each other from high school, where they were all “nerds” who “loved math,” according to Michaelson. The trio was inspired by the idea to create a tool that could personalize recommendations based on consumer tastes, using information like the consumers’ favorite dishes and the “types of vibes” they prefer in a restaurant.
When the trio began the project, it started as a hobby. All three were at different universities, Michaelson said, adding that they eventually started prioritizing the startup over school.
In its next phase of engagement with the technology, the team moved into a space at WeWork in Berkeley to build the first of its two apps. Although neither app functioned successfully, the software built into the two preliminary apps paved the foundation for a powerful profiling software that could predict people’s tastes, according to Michaelson.
The software aggregates data from menus, restaurants and recipes.
“We can start to see, across menus, what things start to correlate. It’s about looking at as many data sources as we can to extract patterns,” Michaelson said.
The software can tailor recommendations to consumers who are vegetarian, vegan or pescatarian. Recipe and restaurant data power the software’s predictions of multi-item combinations.
For example, adding a lime to one’s cart could prompt a vast array of suggestions. However, adding a lime and tonic water would yield a different set of recommendations from adding a lime and a coconut.
The program is precisely tailored to analyze these “unique pairings of ingredients,” according to Michaelson.
The team noted that the types of data powering the software do not involve personally identifiable information — that is, the model preserves the identity of the consumer. Currently, the software is being sold to the top retailers in the grocery industry, according to the team.
Although the current model does not incorporate medical or personal data, the team foresees the software’s potential in the field of human nutrition and medicine.
“We’re going to see a lot of medical applications,” Michaelson said. “You really are what you eat, and down the line, we want to make what you eat better.”
Contact Sasha Langholz at [email protected].