Most of the money in music software right now is chasing the same thing: speed. Type a prompt, get a track. Streaming platforms are drowning in the output. Evan Mezeske, who spent more than a decade leading engineering teams at Google building large-scale systems, decided to run in the opposite direction entirely.
His project is called Anukari, and it is a musical instrument built on a real-time 3D physics simulation. There are no oscillators to detune, no filter envelopes to dial in, none of the furniture that has defined synthesis since the 1960s. Instead you assemble a virtual mechanical contraption out of masses, springs, mallets and microphones, hit play, and let Newtonian physics do the sound design.
Sound as a side effect
The mental model is closer to a science experiment than a DAW session. Anchor a mass, tension a spring, strike it with a mallet, and place a virtual microphone somewhere in the resulting mess of resonance. The tone you hear is not a preset being recalled but the emergent result of forces acting on objects. Move the microphone and the timbre changes. Stiffen a spring and the whole system reorganizes itself.
Mezeske’s team offers a useful mental shortcut: think of stretching a rubber band and plucking it like an upright bass, then multiply that by a million interacting elements. The pitch frames the software as “an invitation to experiment, tinker, and discover something no one has heard before” — a phrase that would sound like standard marketing copy if the underlying architecture were not so genuinely unusual.
Physical modelling synthesis is not new. Yamaha shipped the VL1 in 1994, and Applied Acoustics and Modartt have made careers out of modelled pianos and strings for decades. What is different here is scope and interactivity: rather than modelling one known instrument well, Anukari gives you a sandbox of primitives and lets you construct arrangements that have no acoustic real-world equivalent — and it runs the simulation live, on the GPU, while you play.
Why it matters
The timing is the interesting part. Generative audio models have made “produce a finished song” nearly frictionless, which has quietly devalued the finished song as an artifact. Tools like Anukari bet on the opposite scarcity: not output, but the experience of playing something and being surprised by it. That is a real market — the modular synth world has been growing for fifteen years on exactly that logic — but it is a niche one.
The obvious question is whether physics-driven sound design is legible enough for working producers. Subtractive synthesis is ugly but teachable; a cutoff knob does a predictable thing. A coupled mass-spring system does not. Anukari’s success will hinge less on the elegance of its simulation than on whether the interface can make a chaotic system feel steerable. Mezeske’s Google pedigree suggests the engineering is sound. Whether it is playable is the harder problem, and the one that will decide if this lands beyond the sound-design hobbyist crowd.
Either way, it is a refreshing counterargument in a year where nearly every music-tech launch has been a variation on “the AI writes it for you.”
