Sound, given a shape.
An idea, given a life.

Make sound something
you can see—and work with.

The idea was to turn music into a spatial visual experience. Making that useful meant going beyond an isolated rendering demo: people needed to bring their own audio, shape the result, and connect it to an audiovisual workflow.

That creates a connected product problem. Live audio processing, GPU rendering, user controls, desktop output, and licensing all have to fit into one experience.

Own the whole product.
Build with AI.

I founded Baryon and directed its product design and AI-assisted development. My contribution spans the user experience, system architecture, implementation direction, and verification process.

• Product & interface: defined the experience and controls for exploring and performing with sound-driven visuals.
• Rendering & integrations: developed the GPU-accelerated engine and integrated MIDI/OSC controls and Syphon/Spout output paths.
• Product delivery: developed licensing for trials, device-bound activation, offline verification, and ownership tracking, alongside onboarding documentation.
• Engineering process: built reusable AI workflows for root-cause analysis, focused implementation, review, and regression testing.

AI agents handle much of the implementation. I provide product judgment, sequence the work, direct the systems, and take responsibility for evaluating the result.

One engine.
Different ways to use it.

The browser experience and desktop application share the audio/render engine and application shell. Desktop-specific integrations sit around that shared core.

This keeps the core visual behavior in one place while allowing the desktop app to support native audiovisual workflows. It also creates a clearer boundary for engineering changes: improve the shared engine once, and handle platform behavior where it belongs.

The tradeoff: shared systems need explicit boundaries. A desktop-specific output concern should not become a second owner of the engine’s state.

Generated code is
the start of the review.

Baryon’s engineering process combines unit and regression tests, architecture checks, smoke tests, performance probes, and hands-on product inspection. I use AI-assisted review to question the implementation as well as produce it.

When a defect reveals a recurring pattern, I turn that lesson into a reusable workflow: find the component that owns the behavior, make the change there, simplify the result, and verify the affected experience.

The repository maintains dedicated checks for the engine, application shell, desktop integrations, and licensing. Those checks support development; they don’t replace testing on the target hardware.

A working product.
A repeatable way to build.

Baryon is launched as a cross-platform audiovisual product, with a browser experience and desktop application. It brings live audio, 3D rendering, performance controls, native output integrations, and licensing into a single product.

For me, it is also concrete evidence of the work I offer: taking an ambiguous objective, keeping a complex product coherent, and using AI-assisted engineering to deliver it.

Built with: JavaScript, React, Three.js, React Three Fiber, WebGPU, Electron, and Convex.