AI & Automation · design at scale

The site you’re reading built itself.

Nearly a hundred case studies, one design language, and a pipeline that turned them into this living portfolio— the one you’re in right now.

Not every AI project is a script or an app. My biggest one is a method: I designed ~99 case-study decks in Claude Design, then built the pipeline that renders them as real, linkable web pages in a system I authored. This site is that pipeline’s output.

deck → pipeline → this page

Systematic productionMulti-tool AI orchestrationClaude Design · Drive · Next.jsDesign systems~99 decks → 50+ pages

01The raw material

~99 finished pieces, all speaking one language.

In Claude Design I built close to a hundred projects — case-study decks, hiring presentations, and design systems — all carrying a single authored visual language I call Blueprint: IBM Plex, a grid, a disciplined blue-and-rust palette, light and dark. That consistency is the whole point: it’s what makes ~99 separate artifacts read as one body of work.

But ~99 decks locked inside a design tool isn’t a portfolio. It’s a folder. To be useful, the work had to become a living, linkable, evergreen site.

02The pipeline I built

A repeatable machine that turns a deck into a page.

Instead of hand-rebuilding each study, I built a documented, repeatable pipeline and ran the work in batched “waves” against a locked spec that survives context resets. One proven pattern, replicated across dozens of pages.

Read the deck

Pull each Claude Design deck’s layout and narrative programmatically over DesignSync.

Fetch real assets

Retrieve the actual client screenshots from Google Drive archives over MCP — no mockups.

Offload the heavy work

A background subagent decodes, resizes, and verifies every image, so multi-megabyte base64 never pollutes the main context. Token discipline, built into the process.

Author & verify

Compose each page against a shared primitive library, then confirm every route returns 200.

!The tell

This very site is the artifact.

Here’s the part I love. The case studies about my AI work — the backup system, SaberIQ, running my life on agents — live insidea portfolio that my AI-build method produced. You’re reading the output of the machine, in the machine.

The medium is the proof. If you want to know whether I can use AI to build real things at scale, look at where you’re standing.

03What it demonstrates

Production at scale — with the honesty kept in.

Systematic production

One authored language propagated across ~99 decks and 50+ pages, built as a pattern, not a hundred one-offs.

Multi-tool orchestration

Claude Design → Google Drive → a subagent image pipeline → Next.js → deploy, all driven as one loop.

And the guardrail I’m proudest of: no fabricated metrics. Where a real gap exists, the process leaves a visible NEEDS: flag instead of inventing a number. Speed never bought a lie.

The point

I didn’t just make the work. I built the machine that ships it.

That’s the difference between using AI to generate a page and using it to build the system that keeps a whole portfolio honest, consistent, and alive. You’re looking at the result.

← How I approach AI work