London-based startup Kaikaku.AI has introduced "Epicure," the first trio of artificial intelligence models capable of analyzing ingredient pairings based on traditional recipes or molecular chemical bonds. This project opens up a more scientific approach to culinary creation using machine learning.
Key Developments
To build this tool, the developers trained the Epicure models on a massive dataset of 4.14 million recipes in seven different languages, combined with the FlavorDB chemical database. Each version of the model offers different ingredient pairing suggestions for the same input query, such as finding a suitable side dish for a specific ingredient.
Notably, the model based entirely on molecular chemistry yielded better classification results in terms of superior nutritional value compared to the conventional recipe-based version. Intriguingly, this chemical model had never been directly trained on nutritional data or prior recipes.
Why This Matters
Epicure proves that AI can transcend the limitations of conventional dietary habits to discover unique pairings based on the scientific nature of food. The chemistry-based model outperforming the traditional model opens up significant potential for automating new recipe creation, providing powerful support for professional chefs in Vietnam and worldwide.