About
I'm Daniel Rushton, a senior systems software engineer based in Utah. I spent seven years at NVIDIA on the OptiX ray-tracing SDK: first optimizing CUDA ray-tracing algorithms, then building the test infrastructure that validated them, from stress suites across a global fleet of GPU machines to a performance-history system that caught regressions over time.
Before NVIDIA, I wrote C++ systems software as a co-op at L3Harris and did systems research at the University of Utah, where I co-authored Memshare, a USENIX paper on multi-tenant key-value caching. I've also done graduate work toward master's degrees in computer science at both the University of Utah and Stanford.
And before any of that, I was a recording engineer and producer. I continue to work on music projects today.
Beyond engineering, I have a deep interest in theoretical physics, consciousness, and quantum and biological computing, and especially in where they intersect.
Away from computing, I am deeply interested in the human mind, and mental health. I also enjoy rock-climbing, mountaineering, and working towards my private pilot's license.
Learning GPU kernel engineering, kernel evaluation and LLM evaluation in public, one project and one post at a time. The series and the projects are the record.
Experience
Education
Research
How I use AI
The writing is mine. Every post is in my own words. Claude does the grammar and punctuation passes, and I sometimes use it to edit and fact-check.
The code is written with Claude. Anthropic's AI model writes the code in the projects; I modify, verify and test it before it's published.
So is most of this site. Claude built the bulk of it, and I check it for correctness.
For a series about verifying what AI systems produce, it seemed right to say so plainly.
Get in touch
I'm glad to talk about GPU performance, evaluation, or the work on this site. I also take on a small number of consulting projects: kernel-claim audits, kernel performance reviews and eval harness setup.