Dan Sosa adds new flavors of AI to Parsnip 🫔
A symphony of LLMs, knowledge graphs, and passion for cooking & learning
It's practically unheard-of 🤯 for a new cofounder to join a startup 3 years into the journey.
But, the idea maze and early days of a startup are filled with twists and turns, and while Parsnip’s mission hasn’t changed, our insistent efforts revealed new approaches and new solutions. We began with a north star of solving the cooking problem, and discovered that a huge part of making cooking easy was to make learning to cook accessible. And just as Duolingo created their revolutionary approach to language education in service of a different goal1, we also realized that we would need to build an effective AI-personalized learning engine—something, as it turns out, that would have a far broader impact than just cooking.
And despite riding the requisite peaks and troughs on the startup rollercoaster, we are beyond lucky to have the perfect partner join our team to build this new, more expansive vision. I’m delighted to welcome Dan Sosa to Parsnip, and I’d like to tell you a bit more about him.
An harmonious pairing of AI and gastronomy
Dan has an incredible technical background: PhD in AI / bioinformatics & MS in Statistics from Stanford, and a triple major in computer science, molecular biology, and management (Sloan) from MIT2.
Dan’s PhD research focus was knowledge extraction and reasoning from scientific literature for drug discovery. He was fine-tuning LLMs—before LLMs were cool—to build knowledge graphs from millions of biomedical research articles to find novel uses for existing drugs 💊. As you’ll see, it’s impossible to imagine a more perfect skill set for building Parsnip’s AI-generated skill trees, with knowledge graphs at their core.
Like me, Dan’s PhD forced him to both learn to cook, and learn to learn. As a result, he’s fostered the same passions for learning and cooking, not to mention world-class technical skills, as good as anyone I’ve met.
In his final year, Dan also founded an AI smart cookbook startup with colleagues at Stanford, to pursue a continual passion for food throughout his life. His PhD advisor, renowned bioinformatics professor Russ Altman, put it best when I asked whether Dan really cared about cooking:
Inspired by his biomedical research, Dan thought about how to deconstruct culinary knowledge into a format that would be easy to learn, taking inspiration from his PhD research using LLMs to extract structure from unstructured text. It turns out that the need to organize knowledge is universal3, and the tools from knowledge engineering and ontology development apply straightforwardly to understanding anything—including cooking.
Dan and I finally met in a fateful way when multiple connections who heard about his AI smart cookbook told him to check out Parsnip, including Eric Horvitz, a mutual research mentor of ours.
Most importantly, Dan doesn’t just have one of the best technical skillsets of anyone I've met—he’s also one of most emotionally intelligent people I know.
AI at the intersection of technology and humanity
Since the dawn of Parsnip, we’ve fundamentally viewed the promise of AI not as replacing humans, but empowering them to become the best version of themselves. This idea is central to Dan’s view of the world as well.
An important aspect of achieving that mission is the combination of art and science, our left brains and right brains, in pursuit of a product that not only leverages technology but is also imbued with beauty that deeply satisfies our souls. This isn’t just an important feature of exceptional consumer products, but probably the only way to build them. Apple’s exquisite product craftsmanship had vital roots in Steve Jobs’ fascination for connecting music and art to technology.4
During a PhD program that was daunting for many, Dan nonetheless found time to conduct and compose an original score for the Stanford Symphony Orchestra.5 Using a Raspberry Pi running signal processing algorithms in Python, he also created an LED light show in his bedroom that responded to the changing harmonies of jazz piano improv in real time.
This intersection of art with science, serious deep tech with our innate humanity, is a core way of thinking for both Dan and me6. AI shouldn’t replace human experiences—but augment, deepen, and bring more meaning to them. With music, cooking, and anything else, AI is a paintbrush to be used at the right time in service of our goals and desires, just like any technology that has come before it. We envision the AI behind Parsnip helping us to uncover new solutions and reimagine what’s possible, bringing more meaning and joy to our lives.
People often think of cooking as an art, and indeed on our journey we’ve seen few cooking products that have realized the potential of technology as part of this age-old human skill. But there’s incredible power in bringing ideas from computer science, machine learning, and AI to the way we make our food.
During his undergrad at MIT, Dan became the house mixologist and bartender. He loved the creativity that was possible in making cocktails, and how this would gather people together. Similarly, as Dan revisited his family heritage by learning Guatemalan recipes and delighting his mom, dad, and sister, food again showed its great power to bring people together.
Dan delved even further into mixology during COVID lockdown. His AI cookbook idea came from a desire to understand how harmonious flavors come together in the sublime flavor profiles of cocktails, and to teach the principles of sweetness, acidity, and texture. Rather than taking away our agency, this “knowledge augmentation” allows anyone to get more comfortable with creativity. Parsnip is all about bringing this creative freedom to food, and many more areas of our lives beyond that.
My initial timid attempts at cooking looked a lot like Dan’s: a laughably basic repertoire of a couple of dishes. During COVID, Dan committed to level up his cooking and created his own curriculum to learn all the skills to make an excellent Beef Wellington. Starting with searing meat, he built up to working with puff pastry. He taught himself how to make tarts and chocolate desserts as accompaniments. By the end of this journey, he was able to delight his friends and roommates with a sumptuous five-course meal.
Parsnip is going to bring Dan’s joyful, thrilling experience of leveling up to everyone, except you won’t need to figure out what to learn yourself—AI will do it for you. And once we’re done, anyone in the world will be able to cook anything they want to eat, even if that dish comes from the other side of the planet.
With Dan on the team, we now have all the ingredients we need to cook up our vision of personalized learning, and a new generation of consumer products that empower people to be the best version of themselves. Learn more about that here:
Duolingo emerged from Luis von Ahn’s “games with a purpose” research and work with reCAPTCHA, where users doing one thing (solving CAPTCHAs) would help accomplish something else (digitizing books). Duolingo’s original objective was to aim even higher by teaching users languages while translating the web (I believe this pair of goals is where the “Duo” comes from). Human translation became obsolete because of statistical machine translation and later deep learning (both of which are still trained on human data), but the effective language learning stuck, and here we are!
That’s courses 6-7 and 15 for those of you wearing brass rats!
Google's mission is to organize the world's information and make it universally accessible and useful. We think that “useful” has so much further to go—for example, through the design of a process that is 10 times better at getting that information into your brain 😉.
I highly recommend Walter Isaacson’s biography of Jobs. Isaacson explains that Jobs’ interest in art originated from his father’s taste in the design of cars, and continued through his love of music. This led to seminal innovations in iTunes, the iPhone, and other Apple products. Although Jobs was not perfect by any means, this weaving together of art and science were integral qualities to building a lasting company.
To the mild chagrin of his PhD advisor 😊
Check out some of Dan’s art at https://dansosa.me/art/.
Learning to learn - tha's it!