How I Used Three AIs to Speed Up My WordPress Portfolio

—

Executive Snapshot

FieldProject summary
ProblemSlow real-visitor loading, heavy third-party embeds, and a portfolio that needed clearer paths for recruiters.
System BuiltThree coordinated AI tracks covering performance QA, bulk content operations, recruiter navigation, and an external portfolio assistant.
Tools UsedClaude, ChatGPT, Gemini, WordPress.com API, Jetpack Stats, Google PageSpeed Insights, Custom CSS, and Poe.
Measurable OutcomeHomepage mobile PageSpeed: 81 → 92. Tested Instagram post: 84% smaller. 686 books refactored across 20 shelves. 15 case studies upgraded. Social previews complete across 20 existing portfolio and recruiter pages.
ConstraintsKeep the Carnation theme design and standard WordPress.com hosting.

My WordPress portfolio needed to load faster and make my work easier to evaluate. The real-visitor baseline was a 5.96-second Largest Contentful Paint (LCP) at the 75th percentile. Recruiters also needed a faster way to understand each project’s problem, tools, and documented outcome.

I coordinated Claude, ChatGPT, and Gemini in separate workstreams, assigning each a scope that matched its access and tools. The work covered performance, a 686-book library, portfolio navigation, social previews, and a live AI portfolio guide. It ran within the existing Carnation design and standard WordPress.com hosting.

This followed my earlier 5,968-post archive cleanup. This phase focused on how quickly visitors could load the site, find relevant work, and understand it.

WordPress Performance Results

Mobile test metricBeforeAfterChange
Homepage PageSpeed score8192+11 points
Homepage LCP5.0s3.3s34% lower
Homepage payload710 KiB512 KiB28% smaller
Tested Instagram post score6591+26 points
Tested Instagram post LCP5.7s3.4s40% lower
Tested Instagram post payload2,359 KiB374 KiB84% smaller

Measurement scope: These before/after figures are mobile PageSpeed Insights lab results from the project. The 5.96s baseline is a separate real-visitor, 75th-percentile measurement; a post-change field result has not yet been established. The Instagram figures apply to one tested post. Individual resource savings below are diagnostic figures and should not be added together as a separate total.

Three AIs, Separate Ownership

I set the goals and constraints and coordinated the handoffs. Each AI had a defined edit scope, so the parallel work produced zero edit collisions.

AI / accessAssigned work
Claude — browser and speed QAJetpack Stats and mobile PageSpeed diagnostics; script and font pruning; Instagram embed testing and phased cleanup; image uploads; 390px mobile layout audit.
ChatGPT — direct WordPress APIBulk book-entry refactoring; recruiter entry-page improvements; Executive Snapshots and next-case navigation; internal URL normalization; portfolio knowledge-base compilation.
Gemini — strategy and visual productionDiagnostic review and prompt orchestration; DN site icon; 10 portfolio share cards; external AI concierge architecture.

The division followed the integration layer: browser work for visual checks and performance diagnostics, direct API access for repeatable content changes, and a separate strategy and asset-production track.

Traffic Data Changed the Priority

The initial hypothesis was that the 21-page book-library area was the main performance problem. Jetpack Stats changed that priority.

90-day traffic observationViews
Homepage1,192 of 1,369 top-page views — 87%
Book Library5

The homepage dominated the top-page report. Mobile PageSpeed Insights also identified the LCP element as headline text, delayed by render-blocking third-party scripts and font payloads. That made the homepage’s loading dependencies the first target. The library remained a separate content and page-weight cleanup.

Removing Script and Font Weight

Claude removed the unused Google Analytics / Google Tag Manager load associated with G-K1Y2EBHGYR, saving approximately 160 KiB per page. Native Jetpack Stats remained available for site traffic reporting.

Two lines of Custom CSS also eliminated the unused Inter 500 font file, approximately 150 KiB, while preserving Fraunces display headings and the site’s design.

Across the homepage changes, the mobile score rose from 81 to 92, LCP fell from 5.0s to 3.3s, and the measured payload dropped from 710 KiB to 512 KiB.

Testing a Lightweight Instagram Link

On one Instagram-linked post, the embed carried a reported 1,041 KiB payload. Claude tested replacing it with a plain text link: “📷 View this post on Instagram →”.

The full page fell from 2,359 KiB to 374 KiB—an 84% reduction. LCP improved from 5.7s to 3.4s, and the mobile score rose from 65 to 91.

That test supported phased cleanup of active Instagram embeds on trafficked posts. Visitors could still open the original Instagram post, while WordPress no longer had to load the embedded experience on those pages.

Refactoring 686 Books Across 20 Shelves

ChatGPT used direct WordPress API access to replace 684 Kindle preview iframes with numbered, plain-text lists covering 686 books across 20 published shelves.

The two counts describe different things: 686 book entries and 684 preview iframes. The completed shelves contained zero external iframes or scripts in their book-list content.

Library elementBefore → after
Book presentationKindle preview embeds → numbered title-and-author lists
Markup684 read.amazon.com iframes → plain-text <ol> lists
Published coverageAll 686 books across 20 shelves retained
Example shelvesSpirituality & New Age: 106 titles; Business & Leadership: 56 titles
Links and navigationExternal book links removed; existing intros, category headings, and internal Library navigation retained
Draft shelves7 existing draft shelves left untouched

Each entry followed one format: a bold book title, an em dash, and the author when available. This made the Book Library easier to scan and removed its dependence on external preview widgets.

The earlier Library rebuild established the shelf structure. This pass changed how those books were delivered.

Making the Portfolio Easier to Evaluate

The recruiter path needed clear entry points, consistent project summaries, and an obvious next step. ChatGPT upgraded the three main entry pages: AI Systems, AI Operations, and About & contact.

AreaCompleted change
3 recruiter entry pagesClearer routes into relevant systems, projects, background, and contact information
15 eligible portfolio entriesTop-of-page Executive Snapshots: Problem, System Built, Tools, Outcome
15 eligible portfolio entriesBottom-of-page Next Case Study navigation
8 legacy internal URLsNormalized internal links
2 obsolete or empty stubsAI Model Info and Blog moved to Draft

The result is a more consistent evaluation path: choose a relevant project, scan the evidence, then continue to another case study or contact me. These are completed navigation and content improvements. Recruiter inquiries and conversion lift have not yet been measured.

Lighter Images and Consistent Social Previews

Claude compressed oversized featured images. One example, contentops_ai.png, went from 1.2 MB to 127 KB, with a 52 KB version served on mobile.

Gemini generated a 512×512 DN monogram and 10 custom 1200×630 portfolio cards in navy and cream (#14243B / #FAF3D9). Claude uploaded and cropped the assets, with the cards weighing 41–74 KB each.

The monogram became the site icon and homepage share image. Across the existing portfolio and recruiter set, 20 of 20 pages had social previews at the end of the pass. Claude also audited the layout at 390px mobile width.

A Live AI Portfolio Guide With Zero Added Script Weight

I also launched DeepakNairAI on Poe, linked from the site’s /ask/ page.

ChatGPT compiled deepak-portfolio-knowledge.md, a structured knowledge base covering the portfolio’s facts, tools, documented outcomes, and source URLs. Gemini designed the external concierge architecture, and the WordPress landing page gives visitors a direct route into the assistant.

The assistant gives visitors and recruiters an interactive way to explore my published work. Hosting the conversation on Poe adds 0 KB of chatbot scripts to WordPress; the site carries a normal outbound link.

What This Project Demonstrates

This project combined performance diagnosis, bulk content maintenance, recruiter-focused information architecture, and multi-AI coordination within an existing publishing platform.

The traffic report changed the order of work. A targeted embed test supplied a concrete before/after result. Direct API access handled the repetitive content changes, while separate ownership kept three AI workstreams from overwriting one another.

The measured gains are faster mobile lab loads and lighter pages. The operational gains are 686 simplified book entries, consistent navigation across 15 case studies, complete previews across the existing 20-page portfolio set, and a live external AI guide. The next measurement is whether those changes improve real-visitor LCP and recruiter engagement.

Looking for someone who can diagnose content debt, coordinate AI tools, and improve publishing workflows? Read about my background and get in touch.

Next Case Study: How I Used Gemini and Claude Together to Reorganize 5,968 WordPress Posts.

Browse AI Operations projects, explore AI Systems, or ask DeepakNairAI about the work.