Case Study: Can an AI Be Creative? The Story of GrubMix
- Zoe Goh
- Jul 3
- 3 min read

What's the difference between a master chef and a simple cookbook?
The cookbook can only repeat what's already known. It’s a database, a collection of instructions that have worked before. The chef, on the other hand, can create something entirely new. They can feel the ingredients, understand the harmony of flavors, and compose a dish that’s never existed.
For years, AI has been the cookbook. With our internal R&D project, GrubMix, we wanted to see if we could build the chef.
This is the story of how we taught an algorithm to have a moment of genuine culinary inspiration.
The Challenge: Beyond the Recipe Database
Let's be honest, the problem wasn't a lack of recipes. The internet has billions. The real challenge is that blank, uninspired stare into a fridge holding half a bell pepper, a jar of olives, and some leftover chicken. It’s a moment of creative block, a distinctly human problem that a search engine can’t solve. A search for "pepper, olive, chicken recipes" gives you a list; it doesn't give you an idea.
This was the core of our experiment at SwagSoft Labs. We wanted to push past simple AI that just retrieves information. Our goal was to build a true "generative" model—one that could create novel, useful, and sometimes wonderfully surprising outputs. We weren't building a better search engine; we were trying to build an imagination.
The Solution: Teaching an AI the "Grammar" of Food
You can't teach true creativity with a simple rulebook. Think about it—you don't learn to make jazz just by reading sheet music. You have to immerse yourself in the art form, understand its theory, its history, its feel... and then you learn to improvise. We had to teach our AI the 'grammar' of food.
Here’s how we broke it down:
1. The "Library of Taste"
First, we fed our model a massive dataset of global recipes. But we didn't just have it 'read' them as text. We broke each recipe down into its fundamental components: flavor profiles (salty, sweet, umami, sour, bitter), aromatic compounds, ingredient pairings, and cooking techniques. It was less like reading a cookbook and more like attending every cooking school in the world at once, absorbing not just the what, but the why behind every great dish.
2. The "Spark of Invention"
Once our AI understood the foundational rules of flavor and technique—the culinary equivalent of music theory—we programmed it to creatively break them. This is the heart of the generative model. It can analyze the relationships between ingredients it’s seen before and then extrapolate, proposing a new connection. It could now ask, "What if we used this grilling technique on that ingredient?" or "What if we introduced this unexpected flavor here?" It could suggest a surprising new chord in a familiar song.
3. The "User as a Co-Creator"
Technology this complex has to feel effortless. The final interface for GrubMix had to be incredibly simple, acting as a seamless bridge between human intention and machine inspiration. The user provides the theme (the ingredients they have), and the AI acts as their creative partner, instantly composing a new piece of culinary music just for them.
The Outcome: A Showcase of Creative Capabilities

The success of GrubMix wasn't measured in downloads; it was measured in moments of surprise and delight. It was a successful answer to our initial, ambitious question: "Is this possible?"
And the answer is a resounding yes.
Our "demo reel" isn't a list of features, but a gallery of genuine creation. We saw GrubMix generate things we never could have predicted.
Imagine giving the AI just a few random ingredients and watching it work its magic.
We had some incredible "greatest hits" emerge from these sessions. We gave it 'kimchi, cream cheese, and bacon.' It gave us back the 'Seoul-Philly Fusion Roll,' a baked, savory pastry recipe that didn't exist a second before. We gave it 'salmon, blueberries, and basil.' It generated the 'Summer River Pan-Seared Salmon' with a blueberry-basil reduction sauce—a combination that sounds daring but tastes inspired.
More than just an app, the project resulted in a robust, proprietary engine. It’s proof of our capability to build sophisticated, generative AI solutions that can be adapted for any creative industry, from music to fashion to industrial design.
The Future is a Co-creation

The GrubMix experiment taught us that the most exciting frontier for AI isn't about replacing human creativity, but about augmenting it. It’s about building tools that can be our muse, our co-pilot, our creative partner in the truest sense.
The future isn't man vs. machine; it's man and machine creating something new together. The algorithm can provide the unexpected chord change, but the human touch, taste, and soul are what turn it into music.
Have a business challenge that could be solved by a custom, creative AI? Talk to our Computational Creativity team.
Let's build your 'spark of invention,' together.