About

What I'm doing now

I would rather own a system end to end than hold one piece of it, and that shows up in everything I have going right now. Most of my time goes into Nexus at Tandem Interactive, where the whole agency now runs its client work through software I built. An agency this size would normally rent that tooling and then shape how it works around whatever it rented. Building it instead means the platform answers to how the team actually operates, and it means the AI inside it gets held to a standard we set rather than one that arrived with a subscription. That is the part I am proud of. I spent ten years being the person this kind of software gets sold to, and now I get to decide what it owes the people using it.

The same instinct is why I started writing an operating system. I wanted one that could render its interface for whoever was sitting in front of it, which meant a memory-safe systems language and a microkernel to run it on, and the compiler for that language is now written in the language it compiles. Genos boots on bare-metal RISC-V with no Linux and no C underneath it. It draws real pixels through a GPU queue and sleeps between events rather than burning cycles. What I am working toward is having the kernel meter what any generated code is allowed to cost, because an interface that writes itself is only useful if it cannot run away with the machine.

Generating became its own problem and I solved it the same way. Bobs Studio bundles a native inference engine that speaks twenty-plus model architectures across image and video, and it runs the GPU work in an isolated process that streams latent previews back while it renders. There is usually a checkpoint training behind it. Nepotism ran twelve versions on a pool of roughly 8.5 million images, and the last one logged over two million steps.

How I got here

I started freelancing at eighteen and ran it for about three years, which mostly taught me that work nobody can explain doesn't get paid for. Prodigy Brands took the first real chance on me, and I spent it on product photography and quarterly catalogs for executives who wanted to see drafts before anything shipped. Learning to defend a design to people who were not designers turned out to be practice for every client presentation I have given since.

In 2017 I landed at Apple Insurance as the entire marketing function, where design and copy and the website all came to me and that suited me fine. SEO came out of that as well, because the site needed it and I went and learned enough to do it properly, and that thread has run through every job since. Eleven months at a volume shop called Southeast Publications followed, producing for parks across the US and Canada, where the pace teaches you to stop asking whether a piece is interesting and start asking whether a stranger will understand it in the two seconds they're going to give it.

All of it collapsed into one job at Tabanero Hot Sauce. I did essentially everything visual for the brand, labels and billboards and ads and email and sales sheets and logos, while running the social accounts, shooting the field photography that went on them, and handling their web and SEO work in the same stretch. Make the label and the post and the landing page yourself and you lose the ability to think of them as separate channels. I have planned every campaign since as one conversation rather than five.

Running a department and a full book of clients at Equiturn is where I got serious about measurement. At the peak I had twenty-five accounts going at once and took every meeting on all of them, which meant every number I put in front of someone had to survive being pulled on by people who did this for a living. That is when I started building reporting that anticipated the hard question instead of hoping it wouldn't come. Training the new hires and interns started there too, which I liked more than I expected to.

After years of splitting attention across a client book, Zomee was one brand and all of it mine, and I could finally stay somewhere long enough to watch the work compound. Revenue went from $12M to $25M in eighteen months. The part that mattered most for what came next was the technical rebuild underneath it, because I did enough of that work myself to figure out I'd rather build the thing than write the ticket asking someone else for it.

So in 2023 I taught myself the machine-learning stack, mostly so I could train image models instead of prompting them, and that became Nepotism and then Tenosai. I ran the company for a year, shipped the bot and trained its model, and dissolved the LLC when my cofounder and I split on where it was going. What carried forward was shipping software on a schedule to people who tell you immediately when it breaks.

That is the whole reason the Tandem job works. I build the platform the agency runs on, and a decade of being the person this kind of software gets built for is the qualification that actually matters, because I know which features people open every day and which ones they quietly stop opening by week two.

How I work

Numbers have to survive hard questions. I've sat on both sides of the reporting table, and the fastest way to lose a room is a metric nobody can trace back to anything real, so everything I build gets designed for the moment someone smart pushes back. I also teach while I work, going back to training the new hires at Equiturn, and I still answer the comments on every model I publish, because the stranger reporting a bug is doing me a favor.

What I build with

Everything here is something I've shipped in a real project rather than read about.

Current focus

Essays go up when something I'm building turns out worth writing down. New Studio work goes up whenever a generation is worth keeping, and I take on select consulting when the problem is interesting enough to earn the hours.

Get in touch

Frequently asked questions