Skip to content
Marisa D'AmoreGitHub (opens in a new tab)
← All projects

Web

marisadamore.me AI Project

A research project that turned into this website: an original portfolio concept implemented with an AI coding agent and reviewed for content, usability and accessibility, including screen-reader testing.

  • Active
  • 9 Aug 2026
  • Original concept, content strategy, AI-agent direction, testing and human review
  • AI coding agents
  • Prompt design
  • Content and UX direction
  • Accessibility and screen-reader testing
  • GitHub workflow

This project started as a research exercise - a long look at how a small, independent AI practice might actually operate - and ended up as the site you are reading. The research is what produced the structure; the site is what proved I could direct and deliver it with an AI coding agent.

What I set out to do

Create somewhere to put my work that costs almost nothing to run and does not need a login to publish to. That ruled out a content management system immediately. Everything here is Markdown in the same repository as the code, so a piece of writing and the change that produced it can live in the same commit.

How the AI coding agent implemented it

  • Astro - static output, no server, no adapter
  • Tailwind CSS - design tokens defined once and reused everywhere
  • Markdown content collections - with schemas, so a malformed entry fails the build instead of quietly rendering an empty page
  • Self-hosted fonts - no third-party requests, which means no cookie banner and nothing to consent to

No database. No authentication. No forms. No analytics. Every one of those is something that can break, leak, or demand attention later, and none of them is needed to publish a page.

What it demonstrates

This site is the clearest example I have of how I work with AI tools. It was created through direction rather than dictation: deciding what the thing should be, judging what came back, and rejecting the parts that were wrong.

Two of those rejections are worth naming, because they are the actual skill. The first proposed logo was a set of concentric arcs - which is a wifi icon, and borrowing an existing symbol is worse than being boring. The replacement sunrise-over-a-horizon was worse still: generic stock imagery. The mark that I selected is a prism splitting light, arrived at only after saying no twice.

The same applied to the writing. Anything on this site that claims something about me is either true or marked as unfinished. Nothing here is decorative filler, because a portfolio that overstates gets found out in the first interview.

Accessibility

I checked the site’s color contrast and tested it with a screen reader rather than relying on appearance alone. That process caught two issues no contrast check would have: the sticky header hid whatever element you tabbed backwards to, and forcing wider letter spacing pushed a navigation item off the screen entirely. Both are fixed.

That is the part I would want a hiring manager to notice. Contrast is only one part of accessibility; keyboard behavior, reflow, naming, motion and screen-reader experience also require human review. This was an accessibility-focused QA pass, not a formal accessibility audit.

What I would change

I would document the prompt decisions and rejected directions earlier instead of reconstructing them after the work was complete. I would also begin with a smaller set of finished case studies before expanding the résumé and writing sections. A portfolio becomes convincing through evidence, not through the number of empty categories it provides.

What I took from it

The most valuable part of working with an AI coding agent was not generating files faster; it was learning to give clear direction, evaluate the result and keep factual claims separate from attractive filler. The project also reinforced that good constraints-static output, no database and no unnecessary services-make a site easier to understand, test and maintain.