zzoo.dev — The Invisible Audience

Google indexes your portfolio. Perplexity cites it in answers. ChatGPT pulls facts from it when someone asks about your stack. Or none of that happens — because your site is a visual brochure that machines cannot parse.

Most developer portfolios optimize for one audience: the human visitor. This site is an experiment in serving three simultaneously — humans who evaluate, search engines that rank, and AI systems that extract and cite — from pure static HTML.

Three audiences, one constraint

The constraint is zero runtime. No server, no database, no API routes. Every page is pre-rendered HTML served from Vercel's edge. That sounds limiting until you realize what it enables: sub-second loads, zero attack surface, and a deployment model where pushing to main is the entire workflow.

The interesting question is whether static HTML can carry enough structure for machines without sacrificing readability for humans. JSON-LD schemas give answer engines machine-readable context about every entity — person, project, blog post. An llms.txt file at the site root gives LLMs a plain-text map of the site. Semantic HTML and hreflang alternates handle traditional search. All three channels share the same content; they just read it differently.

Content that shows thinking

Each project gets three tabs. Overview for the five-second scan — what is this, why does it matter. Design for the product thinker who wants trade-offs and rejected alternatives. Engineering for the builder who wants architecture decisions and implementation details.

A hiring manager reads one tab. A fellow engineer reads all three. Nobody wades through content meant for someone else.

All content is file-based MDX with Zod-validated frontmatter. Missing Korean translations fall back to English. No CMS, no database — files in a git repo, version-controlled like code.

Bilingual, not translated

English and Korean exist as parallel content. Same structure, but each written natively in its language. The language switcher is a single button — with two locales, a dropdown wastes a click. New content ships in one language first; the fallback ensures nothing breaks.

Open source

The entire codebase is public. The part worth taking is the content architecture — three-tab project system, Zod schemas, triple-channel discoverability. Fork it, swap the content, deploy.