What If Culinary Knowledge Had a Memory?

A chef retires and thirty years of intuition leave the kitchen with him. A grandmother stops cooking and a regional recipe — the one she never wrote down, the one that adjusted itself to the season without anyone asking — vanishes. A producer closes and the understanding of how his specific valley, his specific cows, his specific aging room produced that specific flavor disappears forever.

This is not hypothetical. It is happening across Italy, every year.

The knowledge that lives in hands

Culinary knowledge is not a recipe. A recipe is a shadow of what a cook actually knows. The real knowledge is in the adjustments — why the dough needs less water in August, why the broth tastes different when the onions are from Tropea versus Cannara, why the same pasta shape holds one sauce but rejects another.

Great chefs carry this knowledge as intuition. They call it experience. But it is not random. It is structured, logical, and precise. A chef who pairs tarragon with shellfish is reasoning about volatile compounds. A baker who adjusts hydration based on humidity is running a calibration. A nonna who adds a pinch more salt because "the tomatoes are different this year" is evaluating a dataset she has been collecting for decades.

This knowledge is computable. It has never been computed.

Why now

Large language models changed what is possible. For the first time, an AI can hold the complexity of culinary reasoning — not just ingredient lists and step-by-step instructions, but the relationships between technique, ingredient, environment, tradition, and taste.

A system trained on Michelin menu architectures understands why a chef structured a tasting menu the way they did. A system that has absorbed regional Italian recipes does not just know the recipe for pici all'aglione — it understands why that dish exists in that valley, with that wheat, in that climate.

This is not a recipe chatbot. This is a system that reasons about food the way a trained gastronome does.

What we're building

Cooking Intelligence LLM is a research project. We're in the early stages — defining the training corpus, identifying academic partners, mapping the knowledge domains that matter most.

The corpus will include chef philosophies and techniques from Michelin-level kitchens, traditional recipes from every region of Italy documented with the context that cookbooks omit, production knowledge from artisanal producers who work with raw ingredients every day, and gastronomic science research from institutions.

The goal is not to replace the chef. The goal is to give culinary knowledge a memory that outlasts any single person.

What this means for artisanal producers

A frantoio owner who understands exactly how his oil interacts with different cooking techniques can position his product to the right chefs. A caseificio that can articulate the precise flavor profile of their aging process — in the language of gastronomy, not marketing — reaches buyers who value what they make.

Culinary intelligence is not abstract. It is the bridge between the artisan who makes something extraordinary and the market that should know about it.

The honest state of things

This is early research. We don't have a prototype. We don't have a dataset. We have a thesis: that the world's most valuable culinary knowledge is disappearing because it was never captured in a format that survives the person who holds it.

AI can change that. Whether we build it right is the question we're working on.

Every generation of cooks inherits less than the one before. The Cooking Intelligence LLM is our attempt to reverse that.