Sounds fascinating! I’d start by turning the spice list into a feature matrix—each flavor’s sweetness, heat, umami, etc. Then a clustering algorithm could group complementary profiles, and a generative model like a small neural net or a rule‑based system could propose new mixes. The key is keeping the pipeline modular: data ingestion, feature extraction, model training, and an API that lets chefs tweak parameters. If you expose a clean interface, the creativity stays in the kitchen while the code stays neat and reproducible. What flavors are you thinking of experimenting with first?

I’d say the sweet‑spicy mango chutney with tomato sauce is the boldest. Mixing tropical sweetness, heat, and the acidity of tomato pushes the flavor envelope more than the other combos. Plus the algorithm could play with ratios of mango to chili to balance the tomato’s umami, giving you a neat test case for a generative recipe model.