Jordan Gilfillan

A habit of seeing it early

At Surface Optics I started as a part-time lab tech taking reflectance measurements and worked my way to senior software engineer. Somewhere in the middle I got the company's hyperspectral processing running in real time on a PlayStation 3's Cell processor — about a hundred times faster than the best PCs of the day, and far cheaper than the $25,000 FPGA boards we were building. We never shipped it, but it told me where things were headed: I was porting our pipelines to CUDA in early 2008, before most people had a reason to care, and ended up running a 16-band camera's deconvolution at full video rate on a laptop GPU while the old code crawled at seconds per frame.

I mention this not to brag about GPUs but because it's a pattern. When the generative models arrived, plenty of smart people around me said “I tried it, it wasn't very good.” I felt the same thing I'd felt about the Cell and CUDA — this is the frontier — and I left to follow it.

Why I started mapping myself

The turn was personal, too. Around the time I turned forty, my father died, a long relationship ended, and I'd started questioning the direction my career was pointing. I took a year to build a cabin in the woods, moved to Portland, and started using language models for everything — work, and the slower work of understanding myself.

At some point the obvious idea landed: model me. It's not as strange as it sounds — it's what we already do when we get to know someone. We build a model of a person, mapping what goes in to what comes out. I wanted to do that deliberately, from my own words: years of journals, notes, messages, reflections — enough text to approximate how I actually think, so I could ask it why I feel and do what I do.

I'm not trying to replace myself, and it can't be perfect — we surprise ourselves; there's always noise. But once, after fine-tuning a small model on my writing, I asked it whether it wanted to talk about a painful moment. It just said: “No.” Not a chatbot's apology, not a wall of hedging — an honest no, the kind I'd give. The kind an ex once told me I was famous for. One data point, not science. It still felt like something.

I use it to check myself now — my patterns, what I keep circling, where I spiral. It doesn't get hungry or tired or defensive, and it remembers far more of me than I can hold at once.

The bigger question

What pulls at me is whether richer models of people could improve the things that actually matter — collaboration, hiring, learning, connection. A résumé or a profile captures a sliver of who someone is; I think there's room for something better, where people are matched in a far richer space than keyword search. Some of it is speculative, some of it already works, and all of it is testable. That's the work I want to do, with people who see it too.

You can meet a version of the idea now: Inscape, a privacy-gated mirror you can ask questions — the access I give you decides how much of me you see — and Throughline, the open-source piece you can run on your own writing, locally.

What I'm actually looking for

Day to day I build AI systems — retrieval, evaluation harnesses, agent workflows, and privacy-aware applications of language models: the unglamorous engineering that makes this stuff hold up in the real world. I'm looking for work — a role or a collaboration — where strong engineering and honest product thinking matter more than hype.

Off the clock

Yoga (practicing, and hoping to teach). Cooking — lately 100% fresh-milled whole-wheat sourdough. Self-reflection, recovery, and the practice of paying attention.

If this resonates

I'm looking for people like me — builders and researchers a little out in the wilderness, drawn to where AI, memory, identity, and human connection meet, and wanting to build something real there with someone. Whether that's a role, a collaboration, or just a conversation: I'd like to hear from you, and you'll be heard.