About
What I built, and what it taught me.
I'm Korey Dillon, an engineer who likes being in the weeds. My path ran from the U.S. Air Force through AI infrastructure to founding LatentAxis, and along the way it became a body of applied-AI work: retrieval systems, document-intelligence pipelines, vision models, and the infrastructure around them. This site is where that work lives now.
I started it because I kept seeing the same gap: real engineering problems solved with generic platforms, or handed to whoever was cheapest. I wanted to build the opposite, systems designed around the actual problem.
Trajectory
- Start U.S. Air Force Data systems
- Then AI infrastructure Systems and tooling
- Then Founder, LatentAxis Applied-AI engineering
- Now Engineer Retrieval, vision, language
Education
- B.S. Computer Science University of Arizona
- M.S. Software Engineering University of Arizona, in progress
Research interests: retrieval, evaluation, and reliable, auditable ML systems. Considering doctoral work in the same direction.
I'm an engineer first. The part I enjoy is the part most people skip: understanding a database schema before touching it, debugging a pipeline at 2am, and shipping something that works in production rather than in a demo.
What building these systems taught me is that the hard part is rarely the model. It's correctness you can verify, failure modes you can see, and handoffs clean enough that someone else can own the result.
Every project got my direct attention. That was never a pitch. It's just how I like to work.
Methods I stand behind
Three principles I've developed across this work and write about in depth.
Compliance-First Architecture
Design for auditability and regulation from day one, not as a retrofit.
Read →Scope Before You Build
Tightly framing the problem is what decides whether an AI project ships or stalls.
Read →Exact Over Approximate
Prefer deterministic, exact algorithms until scale actually forces a trade.
Read →How I work
- In the weeds
- Hands-on through the whole build, architecture down to the last bug.
- Full-stack depth
- Database schemas to React frontends to ML pipelines, not one slice.
- Domain-first
- Learn the problem domain before writing a line of code.
- Built to hand off
- Clean handoffs, no lock-in, code meant to outlive the engagement.
Let's talk
I'm always glad to talk shop, whether it's engineering work, a research collaboration, or something you're building.