AI Alchemy

TL;DR: This article traces my pioneering journey applying computational design and AI to luxury jewelry creation—a field surprisingly resistant to technological transformation. Beginning in 2015 with parametric design and evolutionary algorithms, I discovered how machines could become creative partners rather than mere tools. Despite industry skepticism and dismissal from traditional gatekeepers, these approaches eventually gained recognition in mainstream media and attracted interest from major luxury brands. Today, as generative AI transforms creative fields, jewelry making remains stubbornly traditional in its processes. The industry stands at a crossroads, ripe for disruption at every stage.

My first efforts at serious experimentation with AI began neither with text nor image generation, but rather complex geometry.

In 2015, I was working in the luxury jewelry field, designing unique baubles for high net worth clients. Moving beyond traditional techniques involving pen, paper, or ink, I began to explore computer aided design (CAD). At its far edges, I encountered the subject of parametric architecture, made famous by the likes of Zaha Hadid, but still confined to rare and exotic cases of 'iconic' buildings. Compared to the standard way of creating 3D objects through a series of irreversible commands (move, rotate, shear, twist, etc.), you could structure the entire sequence as an algorithmic flow and then tweak every parameter as many times as you wish. Art blended into technology and veered from sculpting and drafting into matrix transforms and multi-dimensional tree traversals.

Algorithmic representation of a pearl earring

This approach, although unusual, was hardly AI. After all, the operator set out all the operations and dialed in all the parameters. But this was just the beginning. I soon found and began to experiment with evolutionary solvers. The idea was simple: specify which outcome you wish to aim for (e.g., fit a number of pearls within a light yet strong metal covering), connect all input parameters to the solver, and let it create generations upon generations of genes, all competing for maximum fitness. Some time later, a solution emerges that the operator did not intentionally create.

Randomness enters the stage

The role subtly shifted from operator to creator of the environment, an alchemist with the proverbial flask. I could re-run the simulation, altering the conditions or goals slightly and attain significantly different results. As an engineer-artist, I was now co-creating with the machine.

Pearls in a pod: solving pearl encapsulation problem

I found this to be a novel and exhilarating way of creating. While I kept up the “old school” habits of drawing/sketching with pencil and ink, I now had an entirely new toolset to explore a nearly infinite number of variations. Beyond aesthetics, I was able to determine optimal weight distributions (crucial for earrings that need to sit well-balanced on aging earlobes), position faceted gemstones for superior light reflection (important for any colored gemstone that has less-than-optimal faceting, i.e., pretty much anything other than diamonds), create closures/locks/links that fit into the design of the pieces (rather than utilizing whatever stock industry preferred at the time), and generally push the boundaries of jewellery engineering as a solo designer.

Sketching out ideas, the human element

My enthusiasm for these ideas was not met with much reception within the jewelry industry. I recall presenting it to a respected art historian (first mistake: don't present new ideas to people whose job is looking backward) who dismissed me as a quack. Mainstream media changed its tune on AI in early 2021, and I even had a brief spot in The New York Times, where the editor prophesied that soon everyone would become the new Tiffany & Co. Curiously, while smaller players in the industry dismissed the idea yet again as having “little application,” the bigger players like Tiffany & Co. dedicated considerable resources to exploring this budding field.

Combining AI-assisted surface generation with micro-pave setting

Fast forwarding a few years, generative AI has entered the mainstream, first with text and then with image and video generation, followed by 3D Gaussian splatting. And yet, the jewellery field is still playing catch-up, barely incorporating more than surface-level applications, and then just for customer support and marketing efforts. If there is ever an industry ripe for disruption, it's the behemoth of luxury adornments. From generative design (still done largely “by hand” in CAD programs), to additive prototyping, where most models are still crudely positioned on support trees by hand (easily solvable with shortest path algorithms), to casting (slightly more difficult but quite possible to optimize with fluid dynamics simulations), to stone setting and finishing (robotics, machine vision, etc.)—the century-old tradition that always prided itself on innovation now awaits its renaissance through AI. Perhaps what's needed isn't just new technology, but a new generation of designers willing to embrace both artistic heritage and computational creativity, forging a path where craftsmanship and algorithms dance in harmony to create tomorrow's treasures.

Earrings constructed from the principles of Platonic solids