AI Assisted Custom WordPress Portfolio Theme with Dynamic Metabox Architecture

Technical Lead · Full‑stack Developer · Prompt Engineer · QA

📅 2025 (Ongoing)
🏢 Independent (self‑initiated)
💼 In Progress
🔧 Agile · Rapid Prototyping · Iterative QA
Skip to main content

Project Overview

I independently designed and developed a fully custom WordPress portfolio theme using Claude Code for coding and ChatGPT for review, integrating structured meta boxes to make content updates seamless and efficient. The project eliminated the need for external development, achieving 100% cost savings, and improved update speed by 80%, allowing changes in minutes instead of hours.

AI‑assisted WordPress portfolio theme with dynamic meta boxes

Designed, coded, tested, and launched a custom WordPress theme solo using Claude Code for development and ChatGPT as an LLM reviewer. Built a complete content model with custom post types and meta boxes so projects, skills, and timelines are structured, reusable, and fast to update. Established a repeatable prompt‑engineering workflow for spec → code → review → test, enabling rapid iteration without contractors.

Performance Improvement 80% faster
Cost Savings 100% dev cost saved

Challenges & Solutions

Challenges

Building a production‑ready theme from scratch with no external resources
Defining a scalable content model and meta‑box schema that would not paint me into a corner
Ensuring code quality, security, and performance while iterating quickly
Making the UI fully responsive and easy to maintain

Solutions

Created a prompt‑engineering loop: requirements → Claude Code generation → ChatGPT review → refine → test
Implemented custom post types and meta boxes for Projects, Roles, Skills, and Case Studies to enable structured content
Modularized PHP templates and sanitized inputs; added accessibility and Lighthouse checks
Established a lightweight QA checklist and smoke tests for every release

Lessons Learned

Clear, constrained prompts dramatically improve AI‑generated code quality
Define the content schema first; templates and UI fall into place afterward
Small, testable increments beat big drops when working with LLMs
A second LLM as a reviewer catches edge cases the first pass misses