I design AI systems and build the things that run on them.

Most people use AI tools. I build the systems underneath them: personalized systems for individual salespeople, a custom Google Ads server, second brains, and agentic workflows wired into the tools a team already runs. Everything below is real and running, not a deck.

3+ yrs
building systems with AI
22
tools in the AI server I built
3
client systems in daily use

I'm Owen. I started using AI the day ChatGPT launched and never really stopped.

What I actually love is systems and design: taking a messy, manual job and building a clean architecture that does it, then making the thing that surfaces it look and feel right. I went deep on Claude in 2025 and it changed what I could build, from helpful chatbot to full systems that run real work.

I'm not an engineer talking over your head, and I'm not a marketer who just prompts a chatbot. I sit in the middle: I think in both prompts and pipelines, I write without fluff, and I ship.

The stack

What I build with.

AI systems & architecture

Claude Code Claude Desktop Projects & Skills Cowork Agentic workflows Custom MCP servers Connector integrations Browser automation (Playwright) Structured memory / routers Prompt libraries Multi-model workflows ChatGPT

Build & design

Astro HTML / CSS / JS React + Remotion Vercel GitHub JSON-LD schema Responsive UX Photoshop & Illustrator

Marketing & growth

Technical & local SEO Content systems Google Ads GA4 Google Business Profile Performance reporting Positioning & copy

How I work

Document everything Repeatable SOPs Dry-run / safety first Ship, then sharpen Self-directed Fast with ambiguity

Systems

The architecture I'm proudest of.

Real systems with their own structure, not one-off prompts. Some built for people who run on them daily, some because the work was eating my week.

Personalized AI systems Built & in use

A Claude built around one salesperson, wired into the tools their day already runs in.

This is the work now. Most people hear "AI system" and picture a chatbot with a personality bolted on. What I build is a system built around one person: trained on the accounts they carry, the objections they hear, and the way they actually write. Then it gets connected to the software their day already lives in, so it reads the live account record instead of whatever the seller remembered to paste into a box.

  • Trained on how that person sells. Their book of accounts, their objections, their voice. Not a company-wide assistant that knows everyone a little and nobody well.
  • Wired into their real tools. Their CRM, their email, their calendar, their files, the recording of the call they just walked out of. Connected through MCP, so nothing has to be re-typed for the AI to see it.
  • It does the Tuesday work. What changed at a target account over the last year, the brief before the call, the follow-up drafted in the seller's own voice.
  • Safety first. Writes come out as drafts. Billing and personal data stay read-only. Nothing sends itself.
  • Then they own it. One of them took his build into production across six connected tools with jobs running on a schedule, which turned my job into auditing his. Going through it line by line is where I found the broken data path sitting underneath the pricing claims his system had been making.

Claude · MCP & connectors · CRM / email / calendar / drive / call recordings

Custom software Built & in use

A Google Ads server I built from scratch, so an AI can run an ad account like a team.

Not a plug-and-play tool I downloaded. I designed and wrote a 1,460-line Model Context Protocol server in Node.js that connects Claude straight to the live Google Ads and GA4 APIs. Me, paired with Claude, ran a full ad account through it. It's the clearest example of how I think: find the manual work, build the system that kills it.

  • 22 purpose-built tools for campaigns, keyword planning, Quality Score analysis, search terms, geo / device / hour breakdowns, and reporting.
  • Safety first: every tool that writes to the account defaults to dry-run, so nothing changes without a preview.
  • GA4 landing-page data and PageSpeed audits built in, plus domain logic baked into the presets (Ontario geo targeting, healthcare-safe negatives).

Node.js · Model Context Protocol · Google Ads API · GA4 · PageSpeed

Knowledge architecture

A "second brain" that runs a whole business.

I architected a structured Claude system built like an operating system: routers that guide behaviour, an append-only decision log, and a client-installable kit kept clean of private data.

  • Three-zone design: shared core, private operations, and a deployable product kit.
  • Skills that trigger on intent, a session-close ritual, and a single source of truth for voice and positioning.

Claude Code · structured markdown · MCP · memory architecture

Production pipeline

A video engine that turns a script into a finished short.

A React + Remotion framework I built that composes 1080×1920 vertical videos from a script file, with a reusable component library and a documented content methodology behind it.

  • 25-component library: animated diagrams, typewriter code, pop captions, glassmorphism citations.
  • A data-driven scripting system (hooks, loops, pacing) tracked by real retention numbers.

React · Remotion · TypeScript · design tokens

Agentic systems In use daily

When a tool has no integration, I build one.

This is the layer underneath the personalized systems above. The AI has to reach the tools you already live in, or it is just a chat window. Claude reaches most of them through a connector. When a tool has no connector, it drives a real browser instead. When nothing exists at all, I write the server myself. No new software for anyone to learn: the AI comes to the tools, not the other way around.

  • Connectors and MCP: Gmail, Google Calendar and Drive, Zoom, Canva, Figma. If a tool exposes a connector, I can wire it in and have it doing real work the same day. The same pattern extends to a CRM or anything else with an API.
  • Browser automation: when a tool has no integration at all, Claude drives a real browser (Playwright) and does the clicks itself.
  • Custom-built when nothing exists: the Google Ads server above is what it looks like when I build the integration from scratch.

Claude connectors · MCP · Playwright · Gmail / Calendar / Drive / Zoom

real clients, real numbers

Also in my range

I run growth, too. The systems just make me faster at it.

Building is what I love. But I've run the full marketing funnel for real clients, which is where a lot of the system-building came from.

The clinic funnel, end to end

For Lotus Counselling I ran SEO, Google Ads, content, and Google Business Profile, then wrote the monthly report. Organic up 18% month over month and top-3 local rankings on 6 of 12 tracked terms.

Paid search that converts

I plan, build, and run Google Ads campaigns (via the server I built). A recent campaign converted within target cost for private healthcare, with strong CTR and Quality Score.

Strategy & positioning

I've written full growth playbooks (a residential service business's go-to-market, value ladder, and metrics) and positioning for every site I build.

See it live

Want to see any of it live?

Happy to walk you through the Google Ads server, a rep's system, or anything else on this page. One short call.

And if you're a sales rep or account manager, skip the portfolio. Here's the system I'd build for you →

Book a call Connect on LinkedIn
Message me