I'm Liam. For six years I wrote copy that had to convert or get killed. For the last three, I've built the AI systems that scale that work across a $20M+ portfolio. Now those same systems run support, hiring, finance, and operations too.
Every system on this site was built inside Claude Code. Frontends, backends, agents, and the tooling that ships them. One runtime, one builder, real receipts.
Reading in 10 seconds? Everything you need is here.
8 systems. Three continents. One builder.
Production platforms and the internal ops that run the business. Each one built on Claude Code. Tap any row to open the build.
The boring, expensive busywork, automated.
A growing library of real automations I built and ran in n8n (a tool for wiring apps together). Each one takes a painful, repetitive job a team does by hand and turns it into a system that runs on its own, using the tools a business already pays for. Filter by team, and open any one to see exactly how it works.
Every workflow runs on the same backbone, and the guardrail is in all of them: a confidence check with a human fallback, so nothing risky ever happens unattended.
Lead response and qualification
“Leads sit for hours before anyone replies. The hot ones go cold.”
A lead fills in the website form. In one second, AI reads it, decides how hot it is, and either texts them back instantly or starts a nurture email, then saves them to the CRM and pings the sales team. No lead ever waits.
Support triage assistant
“The support inbox overflows with the same questions, and replies are slow.”
Every incoming message gets sorted by topic and urgency, then AI drafts a reply using only your help-centre articles. If it's confident and simple, it sends and closes the ticket. If not, it hands the draft to a human to approve first, so nothing risky goes out alone.
Community moderation
“Your community needs watching around the clock for spam and rule-breakers.”
Every new post is checked against your community rules. Clean posts pass instantly; anything borderline (spam, self-promo, abuse) gets sent to a human with the reason attached. The bot flags, it never bans on its own, and every decision is logged.
Customer research engine
“Teams write copy and make decisions on gut feel instead of what customers actually say.”
Type in one brand, and it listens to what real customers are saying about it in nine different places: Reddit, Amazon reviews, TikTok, competitor Facebook ads, and more. Then it checks it gathered enough, and notes exactly where every quote came from, all before anyone writes a single word.
New-customer onboarding cascade
“A sale closes and six teams scramble to do their bit by hand.”
The moment a sale is marked “won,” six teams spring into action at the same time: the invoice goes out, onboarding tasks are created, the customer's account is set up, welcome emails start, and leadership gets a heads-up, all without anyone lifting a finger.
HR handbook assistant
“New hires flood HR with questions already answered in the handbook.”
A new hire asks a question in Slack. The bot looks it up in the employee handbook, answers only from what it actually found (and links the page), and if it isn't sure, it quietly hands the question to a real person. It never makes anything up.
Built, wired and tested end-to-end on my own n8n. The plain-English names are what you see; the real APIs and models sit one layer down in each step. More being added.
What the systems actually produce.
Real static ads my systems generated, each routed through a different expert framework and composed around the brand's real product. Tap any one to enlarge.
Generated by my creative systems, not hand-designed. Every product is composed around the brand's real packaging, accurate to the SKU.
The ads, and the thinking.
Static ads my systems generated, composed around each brand's real product. Every one is a deliberate call — the angle, the psychology, the layout. Tap any ad to read the reasoning behind its copy and its design.
Generated by my creative systems, not hand-designed. Every product is the brand's real packaging, accurate to the SKU — the system composes around it, never invents it.
AI video ads, start to finish.
Full video ads generated end to end with AI: script, voiceover, footage, and edit, each built around the brand's real product. Tap any one to play with sound and see exactly how it was made.
Each ad below leads with a different technique: lip-sync, b-roll, reengineering, animation. The pipeline underneath is the same every time, and so is the rule: the product is never invented. Every ad starts from the brand's real packaging and is composed around it, accurate to the SKU.
Generated end to end with AI: script, voiceover, footage, and edit. Every product is composed around the brand's real packaging, accurate to the SKU.
Real numbers from production systems.
Every number below ties to a system I shipped. No vanity metrics, no rounding tricks.
From copy to code, across three continents.
Six years of direct response craft. Three years of production AI systems. The line between writing and shipping got thinner every year.
DR copywriter → Head of AI Copy
Started here. Daily emails, push notifications, UX writing, and ad copy for a crypto gaming platform with 1.5M+ players. Grew from daily direct response into Senior Copywriter, then Head of AI Copy, building Claude Code agents for the Affiliate and VIP teams. Five years of copy that had to convert or get killed.
AI Copywriter → Head of AI Copy
Joined as the first AI-focused copywriter across the Foxelli portfolio, with Hooks & Needles and Mindful Souls as my primary brands. Six months in I was running Claude in production, not as a drafting tool but as the actual output. Promoted to lead a 5-person copy team through the AI transition, then to take AI across the whole company.
Head of AI Automation/Strategy
Owned AI strategy across 7 DTC brands and a $20M+ combined portfolio. Built Foxelli Studio, an end-to-end AI ad production platform. Built copy agents deployed company-wide and research systems that turned customer surveys into AI-ready brand context. Architected for 100 concurrent users.
Founder
Custom AI marketing systems for DTC and CPG brands at $10M–$100M. Five-layer architecture: Context, Data, Intelligence, Agents, Commands. The team stays in the chair. Every output gets reviewed. The system gets smarter with every correction.
Five principles that decide what gets built and what doesn't.
Claude Code is the build environment, not the assistant.
Every system in this portfolio shipped through Claude Code. Frontends, backends, agents, custom tooling, CLIs. One runtime. Most people use AI to write code. I use it to ship systems.
Humans where it counts. Autonomous everywhere else.
Anything that touches a customer gets a human's approval, and brand voice and creative instinct stay human. Back-office systems run fully autonomous, no approval loop. The real skill is knowing which is which.
Every system gets a learning loop.
If it can't get smarter from the work, I haven't finished building it. Editor feedback, customer corrections, failed campaigns. All of it feeds back into the next pass.
Build the meta-tool.
Don't write a script when you can write the skill that writes the script. Don't write the skill when you can write the agent that picks the right skill. The leverage compounds.
Run AI economically.
All media generation routed through Kie.ai at ~60% under retail. Token discipline upstream: prompt caching, tight context windows, reuse over re-fetching. Cost-engineering an AI stack is engineering. I do both halves.
The runtime, the surfaces, and what's wired in.
Most portfolios list AI alongside React. Mine doesn't. AI is the layer everything else runs through.
The runtime, the surfaces, and what's wired in. Most portfolios list AI alongside React. Mine doesn't. AI is the layer everything runs through.
Runtime
- ▎Claude Code (primary)
- ▎Claude Agent SDK
- ▎60+ custom skills authored
- ▎Custom hooks + settings.json automation
Surfaces
- ▎VS Code (daily driver)
- ▎Google Antigravity
- ▎Terminal CLI
MCP servers wired
- ▎Supabase
- ▎n8n
- ▎Notion
- ▎GitHub
- ▎Firecrawl
- ▎Figma
- ▎Stitch
CLIs addressable
- ▎Codex
- ▎Gemini
- ▎NotebookLM
Verified by the people who built the tools.
Anthropic certifications, English fluency, and the receipts behind the work. The certs come from the lab whose tools I build on every day.
Let's build something that compounds.
Open to senior AI engineering, AI strategy, and advisory roles. Remote · US & EU only. Available now.