This site is small on purpose: a handful of pages, no framework, and nothing to install. I rebuilt it from scratch after the old host wiped it. Here is what it runs on, how the pieces work, and where AI fit into the build.
This page is in English for now.
Every page is plain HTML, CSS and JavaScript. There is no build step and no framework, so any browser can open the files straight from a folder. It is hosted on Cloudflare Pages, which serves it from a global network for free.
I chose this after losing the first version. The old site lived inside a sitebuilder plan, and when the host removed access, the site went with it. Plain files can move to any host in an afternoon. More on that in Notes.
The look borrows from logs and dashboards: status lines, coordinates, version numbers, monospace labels. The idea is that my site should read like the systems I run, not like a brochure.
Section labels carry English and Spanish side by side, and dotted labels reveal their Spanish on hover or keyboard focus. The EN / ES switch in the menu translates the Home, About and Contact pages in full.
The HD on the homepage is not an image. The script draws the letters once on a hidden canvas, samples which pixels are filled, and turns each sample into a small square that eases into place. Move your cursor across it and the squares get pushed away, then settle back.
The first version rendered as a scattered mess. The bug was a fractional canvas width: the browser reported something like 394.4 pixels, and the math that reads the pixel data assumed whole numbers, so most samples landed in the wrong place. Rounding the width fixed the sampling. At that narrow size, round dots still read as noise, so I switched to tightly packed squares, which rebuild clean letterforms.
If your system asks for reduced motion, the letters draw in place with no animation.
Ctrl K (or ⌘K on a Mac) on any page to jump anywhere, copy my email, or grab the resume.I split the work across layers. ChatGPT is the thinking layer: strategy and pushback on decisions before I commit. Claude is the execution layer: the design system, the screens, and most of the code, working from my specs and decisions. I own the direction, the copy, every number, and the call on what ships.
The rule that made it work: nothing important lives only in a chat. Decisions, verified facts and the approved wording for every metric sit in my own documents, which is exactly why this site came back in days after the old host wiped it.
This is the same way I approach AI operations work generally: a layer that questions the plan, a layer that executes it, and a paper trail a person can audit.