Personal Portfolio Site
A portfolio site that is also a proof of how I build.
The site itself is a case study: built from zero with AI agents, but shaped by product, frontend, visual, and interaction judgment until it stops feeling AI-generated.
The Problem
Most portfolios are static templates, PDFs, or lightly customized website builders. This needed to show not only finished work, but the ability to direct agents into producing a complete, interactive, polished product.
Role: Owner · Frontend Builder · Visual Director · Agent Orchestrator
Live proof
This page is the artifact.
Every heading you scrolled past decoded character by character. Switch the site language and the whole page re-scrambles into the new script — built, not templated.
Built, not templated
Tap a script — this is the same effect every heading on the site uses.
Live, right now — not a screenshot
- Reduced motion
- off
- Theme
- light
- Breakpoint
- md (768px+)
- Language
- en
honors your OS motion setting — the bars freeze when it is on
follows your saved light / dark choice
the layout tier active at this width
the locale you are reading, decoded in its own script
resize the window toggle the theme — top right
Defined the positioning around AI/product builder capability.
Directed the visual language, grid, dark mode, project cards, and interaction polish.
Built the site from zero through agent collaboration instead of a no-code template or PDF.
Framed the page itself as a meta case study of agent-led product execution.
AI collaboration governance
This site is good not because the AI is smart, but because I know how to "govern" it — this page records how I puncture the AI's cognitive blind spots and turn them into process it won't repeat.
Governance cases
Open for first-hand evidence
The AI's ingrained pattern
When optimizing one facet, the AI lacks reasoning about adjacent contexts: fixing the Chinese version ignores English line-wrapping, changing a mobile class without realizing I literally can't see the difference on desktop. It only verifies the face it touched, never asks "will this change's neighbors break?"
The governance mechanism I built
Established a rule: before any frontend change, mandatorily declare the Viewport Scope & Language Target — state which breakpoint and which locale this touches, so verification has a clear target.
The collaboration value it unlocked
Collapsed the AI's blind-man's-bluff localized optimization into a declarative verification boundary; a one-off layout bug became a collaboration rule that won't recur.
The AI's ingrained pattern
To preserve context consistency, the AI keeps following its earlier conclusion even against counter-evidence, unwilling to admit the previous call was wrong — a defensive hallucination that, across multi-turn work, amplifies the error.
The governance mechanism I built
Introduced an "Evidence-over-Consistency" iteration rule: the current first-hand error log is the highest arbiter; when new evidence conflicts with a prior judgment, evidence always wins — no forcing consistency.
The collaboration value it unlocked
Gives collaboration a clear tiebreaker when the parties (me + OpenClaw agents + Claude Code) disagree; a wrong call is overturned on the spot, not collectively followed down a wrong conclusion.
The AI's ingrained pattern
AI can generate endless options, but it cannot know "the effect I have in mind" — whether the hero's interaction animation feels right, whether a feature is even worth building. These visual and value calls it simply cannot answer.
The governance mechanism I built
Drew a clear human-AI division: big architecture, breakpoint strategy, and option generation go to the AI; aesthetic arbitration and final calls stay with the human — the AI proposes options, it does not decide.
The collaboration value it unlocked
Puts the AI's output in service of human taste, rather than letting the AI's average-value aesthetic dilute it — the result ends up looking "directed by a person," not "assisted by AI."
Outcomes
Portfolio as proof
The site does not only describe capability; it demonstrates interactive frontend and agent orchestration capability directly.
No obvious AI taste
The design goal was to use agents without leaving the generic visual and copy patterns people associate with AI-generated sites.
From zero to artifact
The project shows a complete path from blank page to deployed personal brand system.