Back when Kai Schaeffner was running Thermomix North America, he helped build what became one of the most successful digital recipe and connected appliance platforms in the U.S. and beyond.
Still, despite that success, Schaeffner often wondered whether he and the appliance industry had the right focus. Were they spending too much time trying to create a connected device ecosystem and not enough time developing compelling features that actually made consumers’ lives better?
“The big question was, okay, now I’m connecting all my appliances in the kitchen and what’s the value of it? And I think the industry still owes the answer,” Schaeffner said.
At the time, he suspected that the value proposition most likely to resonate with consumers, at least in the kitchen, might be better taste and nutrition. But it was still early, and he could only do so much while trying to build the Thermomix business in North America.
Schaeffner eventually left Thermomix in 2022 and began consulting, but he never stopped thinking about how technology could improve taste and nutrition for the home cook. Before long, he met a like-minded technologist in Andreas Nicklas, a longtime computer scientist who had spent much of his career applying advanced technologies such as AI to healthcare.
Nicklas had worked on AI models designed to address long-term health issues such as diabetes and, through that work, came to a realization similar to Schaeffner’s, only by a very different path.
“What I found out is that people do not accept healthy food if the food doesn’t taste well,” Nicklas said.
Both native German speakers, Schaeffner and Nicklas held their first meeting, where else, at the opera in Zurich. They hit it off immediately.
“I think we were three, four hours sitting there because we could not stop sharing the ideas,” Schaeffner said.
They soon began discussing how they might work together, with a focus on the intersection of taste and health. The more they talked, the more they became convinced that AI could help steer consumers toward healthier food without asking them to sacrifice enjoyment.
“It is a journey where both sides keep on learning and that’s the beautiful piece of AI,” Schaeffner said. “Because you will learn what’s good for you and the reaction of the food you’re eating and the impact you’re understanding, the knowledge you increase, and each feedback you provide.”
The company they are now building, called Jaicara, uses not one AI model but an ensemble of specialized models trained on scientific cohort and patient-study data related to human taste and smell.
Before they could train the models, the team first had to clean and structure the underlying data because many of the studies had not been organized for machine-learning applications.
“Once you have analyzed the data, we usually build some first AI models based on the stuff we have available to find out how strong the signal is in the cohort or the patient studies in the data,” Nicklas said.
Once the team determines that the data contains a strong enough signal, it develops models for the particular application.
“Once we are sure that for a specific intent, the signal and the data is strong enough, we then start to build specific AI models for the specific purpose of the project,” Nicklas said.
According to Nicklas, a consumer using the platform through a partner such as Thermomix would take a roughly 20-question online assessment. The system would use the answers to infer that person’s taste and smell profile, which could then be matched against a recipe database to identify dishes the user would be more likely to enjoy.
The two emphasize that they built proprietary models rather than relying on large language models, citing health-data privacy requirements, the need for specialized predictions, and concerns about hallucinations. Nicklas said the company adapts its models for different cohort studies, patient studies, and healthcare applications rather than using the same model for every purpose.
One potential application the two foresee is a digital food passport that could allow a restaurant or hotel to understand a customer’s preferences and recommend suitable dishes.
“If people come into his restaurant and he knows their digital taste and smell preferences, he can perfectly serve his customers or even create specific dishes for them,” Nicklas said.
This summer, the two are beginning discussions with potential partners as they look to move beyond their initial proof-of-concept work.
To hear my full conversation with Schaeffner and Nicklas, click play below or find the episode on Apple Podcasts or Spotify.