For the second Computational Creativity assignment, we developed a generative AI system using genetic algorithms to create novel cookie recipes by analyzing and combining elements from existing recipe data, then compiled the results into a custom-designed cookbook. I was responsible for dataset curation and annotation, visual content generation using ChatGPT, Stable Diffusion and Photoshop AI, the cookbook design, and actually baking the final cookies, while my teammate handled the genetic algorithm implementation using the SimplePIERRE framework in p5.js. The challenge was to see if computational creativity could generate genuinely interesting and edible cookie recipes that went beyond typical combinations, including both sweet and savory options worth actually baking.

I curated a balanced dataset of 20 recipes (10 sweet, 10 savory) and developed a five-category ingredient classification system (neutral, sweet, savory, bitter, sour) to guide the algorithm's decisions. The most tedious part was standardizing measurements from different international sources, converting between US cups and metric systems using online conversion tools to ensure accuracy. Working with AI generation was both hilarious and frustrating. The genetic algorithm would sometimes create recipes calling for 200+ eggs due to crossover mutations, and ChatGPT would flag these as "challenging cookies" or advise eating them "at your own discretion." 

For the cookbook design, I chose a coding-inspired aesthetic using programming syntax elements throughout the layout, which perfectly bridged our computational approach with traditional cookbook presentation. Each recipe page included AI-generated food photography (created using the recipe titles as prompts), detailed taste profiles, cooking instructions, and clear iconography. We delivered a cookbook featuring 10+ AI-generated recipes and tested our system's viability at our lecturer's "Great Bake-Off" competition, where we baked our algorithm's "Double Chocolate Pepper Marshmallow Cookies" and won the "most surprising" prize. This project taught me that data structure and classification directly impacts AI creativity, how you organize and annotate training data determines whether you get interesting culinary innovations or just random ingredient chaos. The experience of taking a computationally-generated idea and proving it works by actually baking it was incredibly satisfying and demonstrated the value of combining theoretical AI creativity with real-world validation.