Adversarial ML researcher with a Computer Vision focus. I treat Swift — on the device and on the server — as the medium for machine learning and creative solutions.

For years I worked in computer vision as an applied engineer alongside researchers: porting their models efficiently onto edge devices and shaping the inputs and outputs they trained against to hit competitive benchmarks. On paper my career has been iOS / iPadOS / macOS engineering — in practice my work has always lived where creative engineering meets ML systems.

I'm now turning that toolkit toward Adversarial ML research, still in Swift, still close to the edge — with one question running underneath most of what I build:

"When does generative AI qualify for fair use?"

Frigate runs the on-device ML MLX stack that powers various projects I am building. I am developing an entire stack from text to images to audio with royalty tracking infrastructure. To merge the world of creation and generation faithfully and with accountability. There are sub-projects that sprout from these 3 core pillars at points in time. For instance "Bonnie" from audio ("Synth") related findings, "Gita's Wish" from text ("Seer") related findings, or "Obscur" from image ("Garde") related findings. Some core repos are not public until it is the proper time legally and functionally.

Can 1 man really pull this off? Will this lead to a new management style for future generations, if so? Writing technicals of each facet of the project will live in a blog on my site. To back the understandings of these projects, a concrete bridge between generation and understanding in a complex engineering world where everyone is trying to re-discover the value they bring to the table.

I do believe 2026 to be the year of the strategic inflection point. The final vision that unifies all of my work will be unveiled in 2027.