Building a Meadow: Why My Portfolio Isn't a Website

Most portfolios are filing cabinets. A hero section, a projects grid, a contact form. Functional, forgettable, done in a weekend.

Mine took longer than the actual projects it showcases. That should have been a warning sign. Instead it became the point.

The premise

I didn't want a site. I wanted a place somewhere a visitor could walk through, not scroll past. So I built a meadow. Not literally, but structurally: seven pages, each a "location" in one continuous world, connected by the same light, the same ground, the same rules of physics.

The idea sits at an intersection I keep returning to in everything I build: Ancient × Nature × Engineering × AI. Old forms roots, stone, seasons carrying genuinely new machinery underneath. A terminal that feels excavated rather than launched. A chatbot that lives in the underbrush instead of a corner widget.

Five colors, one clock

Restraint was the hardest part. Early versions had a dozen accent colors and I hated all of them. I cut it down to five tokens cream, ink, moss, amber, bark and forced every screen to live inside that palette. Constraints like this don't limit a design, they are the design. Once the palette was fixed, decisions that used to take an hour took thirty seconds.

Then I added a clock. The whole site shifts through dawn, day, dusk, and night based on the visitor's local time. Not a toggle an ambient, unannounced thing. If you visit at 11pm, the meadow is dark and quiet. If you visit at noon, it's bright and busy. Nobody has to notice for it to be working.

The terminal that thinks it's a root system

The centerpiece is what I call Oracle Root a command-line interface styled like something you'd dig up rather than boot up. It uses a command registry pattern under the hood (nothing exotic, just clean separation of intent and execution), but the skin is the point: a root sigil as the trigger icon, organic motion instead of blinking cursors, copy that sounds like it's speaking from soil rather than silicon.

Underneath it sits a RAG-powered chatbot Gemini 2.5 Flash with hybrid BM25 and vector retrieval that actually knows the site's content and can answer questions about my work. It's not a gimmick bolted on for the "wow." It's meant to be the fastest way to get an honest answer if you don't want to read seven pages of meadow.

The filter that cut the most work

Somewhere in the build I noticed I was adding things because they were cool, not because anyone would use them. So I made a rule: every feature has to survive one question who reads this?

A lot didn't survive. A parallax layer that existed purely to exist. A second easter-egg terminal command nobody would ever type. Aesthetic flourishes that served my own delight more than a visitor's experience. Cutting them didn't make the site less distinctive it made the distinctive parts land harder, because they weren't competing with noise.

Why bother

A filing-cabinet portfolio tells people what you've built. A place tells them how you think. Anyone hiring, funding, or partnering with me is really asking one question: does this person's judgment hold up under their own scrutiny? A meadow that's internally consistent same palette, same physics, same restraint, page after page is a much better answer than a bullet list ever could be.

It's slower to build. It's harder to finish. It's also the only kind of portfolio I'd actually want to walk through myself.

This is the first post in what I'm hoping becomes a running log of how Oracle Root, Canopy, and the rest of this world get built. More soon.