Experiment 01

Building My Portfolio with AI

This experiment documents how I used ChatGPT and Codex as collaborators to plan, build, critique, refine, debug, and launch my portfolio. AI helped accelerate the process, but the direction came from design judgment: what to show, what to protect, how the site should feel, and what each section needed to prove.

ChatGPT

Copy, review, critique

Codex

Build, refine, debug

Figma

Design judgment and assets

GPT-5

Model used in Codex

Approach

A live product mindset.

ChatGPT helped me clarify the story, sharpen the positioning, and think through how to explain the work in a way that feels honest and credible. Codex worked inside the actual codebase, which made the process more concrete: ideas could become pages, components, routes, styles, and fixes.

The important part was the loop. I would describe a design concern in plain language, evaluate the output, and keep refining until the portfolio felt more intentional. AI made the workflow faster, but the final decisions still depended on taste, hierarchy, content judgment, and user experience thinking.

Workflow

Collaboration flow

Planned before building

I used AI to clarify the goal, section structure, content direction, and implementation path before making changes. This kept the work focused instead of random.

Switched models based on the task

I used stronger reasoning for planning and content decisions, coding-focused support for implementation, and faster iterations for small visual refinements.

Created reusable Skills

I used repeatable AI workflows and instructions to make portfolio work faster, keep outputs consistent, and avoid explaining the same context again and again.

Built, reviewed, refined

Codex and ChatGPT helped move quickly, but I reviewed the output like a designer: hierarchy, spacing, clarity, tone, privacy, and final user experience stayed human-led.

Proof Points

What changed with AI

Faster IA decisions

AI helped test multiple navigation structures quickly, so I could choose the clearest visitor journey. The final IA reduced guesswork and made key sections easier to find in fewer clicks.

Stronger visual consistency

AI surfaced inconsistent type patterns across pages, then helped apply shared heading and body rules. The result is a portfolio that feels cohesive while keeping strong hierarchy where it matters.

Better mobile scanability

AI accelerated mobile QA by catching spacing and density issues section by section. Tightened gutters and card rhythm made the content faster to scan and reduced unnecessary scrolling.

Safer case-study sharing

AI helped implement selective access logic without breaking the main browsing flow. Sensitive case studies stayed protected while public work remained discoverable for recruiters and collaborators.

What I learned

AI still needs judgment.

Used AI for

PlanningCopywritingCode generationDebuggingDesign critique

Learned

AI is great for momentum, but still needs strong direction, taste, and review.