In this episode of the Sunny Ray Show, host Sunny Ray speaks with Cat, a chemist turned inventor turned founder who built Tastry, a company creating a first principles AI model for human taste, smell, and texture preference. Cat explains how her chemistry background and a master's in bioinformatics led her into the wine industry, where she noticed that mapping words to molecules failed to predict what people actually enjoy. She describes uncovering that identical wines sold under different labels received wildly different critic scores, and how Tastry's model eventually helped winemakers salvage crops during California's wildfires by solving smoke taint issues before harvest. The conversation covers why taste and smell remain modalities AI has largely ignored, Cat's skepticism toward AGI hype and large language models, and how Tastry is expanding beyond wine into skincare, beverages, and CPG. She also reflects on raising almost fourteen million dollars, choosing investors aligned with deep tech timelines, and what it takes to bring specialized AI into traditional industries.
A chemist turned founder explains how Tastry built an AI that predicts human taste, smell, and texture preference before consumers know what they want.
What is Tastry in your own words, and why should people care?
Tastry taught an AI how to taste. It is a human modality AI hasn't tackled, and as a first principles model it applies across the supply chain, from making and recommending products to marketing them and solving supply chain inefficiencies. We started in R&D first, but we are building a platform that touches nearly every part of the process.
What did your experimental streak look like in chemistry before Tastry?
I was pursuing a master's in bioinformatics while working as a chemist to pay for school. I was always too experimental, obsessively reading research papers and poking holes in them. Working in the wine industry, I realized mapping words to molecules wasn't adequate. What mattered wasn't how people described taste, but how much they liked it, and that turned out to be predictable through chemistry.
Tell us about the two labels moment, one identical wine with wildly different scores.
There's a dirty secret in wine: you can make one large commercial batch and sell it under two different labels, like the same product in different cans. We had data showing critics unknowingly rated identical wine differently depending on the label, while consumers were far more consistent. Critics have other incentives at play, which makes their scores less predictable than we'd assume.
Why are taste, smell, and texture still missing modalities for AI?
Other data sets have a latent variable, and for us that variable is chemistry. You cannot buy the data you need or use typical LLMs to crack taste, you need high quality data and specialized algorithms built from first principles. Other sensory AI companies failed because they didn't approach it that way. You have to be expert in chemistry, AI, and consumer science all at once.
What did the first winery results feel like when the model predicted what people would love?
There was a lot of skepticism since wine is a traditional, insular industry. Telling winemakers I could cut six months of iteration down to two hours with 93 percent accuracy sounded outrageous, so our entry point was solving emergencies, not claiming we'd make the best wine. During the California wildfires, our AI helped address smoke taint before growers had to drop entire crops, and that made people believers.
You're skeptical of AGI hype. Why does a failure like an LLM breaking on a swapped chess piece matter so much?
I think the market has confused generality with superiority, mistaking a scaling curve for a theory of intelligence. Studies going back to 1988 show statistical pattern learning isn't the same as understanding language, and a 2023 Berkeley study found children can do abductive reasoning that AI still cannot. I believe the future is an ecosystem of specialized narrow algorithms, not one omniscient general intelligence.
What has building Tastry taught you about capital, when to raise and when not to?
We've raised almost 14 million dollars, though I wouldn't say I've mastered fundraising. I've learned to identify which investor camp fits us: fast growing AI wrapper companies with high churn, or long childhood, capital intensive companies that take years of R&D but become defensible once they mature. We deliberately chose to align with the latter camp.
What do you want deep tech founders watching this to take away from your story?
You shouldn't compare yourself to non-deep tech companies, because deep tech requires a different mentality around timelines, funding, and go to market. It is a longer and harder road, but that's precisely because the payoff is bigger and more durable.
Building something daring? Sunny talks to founders like this every day. Fifteen minutes to see if your story belongs on the stage.
Claim your pre-interview