I don't write most of my code. I direct the agents that do.
I'm a marketing operator by training (BComm, Guelph) who ships production software with AI coding agents — Claude Code daily, Claude API + MCP in everything. No CS degree, no engineering title, and every system below is live, solo-built, and operated by me: deploys, monitoring, billing, and the failure modes at 2 a.m. If you're hiring people who ship with agents rather than despite them, that's the whole profile. Available now for full-time remote roles — open to contract.
Live systems — one person, agent leverage
JARVIS — proactive life-OS PWA
A personal operating system as an installable PWA: natural-language capture, AI-generated daily briefings, cross-device sync, push nudges on cron, and Google Calendar auto time-blocking. The hard part wasn't features — it was reliability, like diagnosing a push pipeline that conflated permission with subscription and silently swallowed failed sends, then shipping self-healing re-subscribe. I use it every day, which is the most honest QA there is.
Foreman — autonomous product-health system
A heartbeat that health-checks my portfolio of live products every 3 hours — deploy rot, paused databases, auth gaps, funnel metrics — and turns anything needing attention into one-tap action cards with pre-written payloads, pushed to my phone. Battle scars included: scheduled tasks silently skipping on battery power, offline runs painting false all-red alerts, and a dueling-systemd-unit restart loop on a remote agent VPS (656 restarts) traced and killed.
LinkRescue — affiliate-link integrity SaaS
Network-agnostic link-rot monitoring for affiliate publishers, including real crawler infrastructure: time-budgeted scans with graceful partial results, parallel link checking with a concurrency-safe domain limiter, and a tiered honest-bot → browser-profile retry strategy. Its monthly data study checks ~6,500 links across 50 sites and feeds a published index — the marketing engine and the product are the same machine, which is very much the point of me.
linkrescue.ioAI Commerce Radar — AI audit funnel
Paste an Etsy/Shopify listing → Claude grades six visibility dimensions and simulates buyer queries against the listing → a shareable scored report with dynamically rendered OG cards, email capture, and checkout. Designed as a growth loop first and a product second: every report is its own distribution.
ai-commerce-radar.vercel.appMCP servers & a working CRM
Published MCP servers — SelfHeal (self-healing proxy for unreliable APIs: retries, circuit breakers, observability), SiteHealth (full website audit, zero API keys), LeadEnrich (waterfall enrichment) — plus a production CRM that runs a real painting business: quotes, invoices, lead pipeline, mined from municipal permit data.
The method, stated plainly
The agent writes the code. I own everything else.
"Everything else" is the actual job: decomposing the problem, deciding the architecture, catching the agent when it loses the plot, verifying end-to-end, and operating what ships. I run multi-agent and headless pipelines with budget caps, keep a shared working-memory protocol so parallel agent sessions don't collide, and monitor production because I've seen what happens when nobody does.
The marketing background isn't a caveat — it's the other half of the leverage. I build the funnel and the feature as one system: the data study that feeds the blog that feeds the scanner that feeds the upgrade path. Software that markets itself, built by someone who's done both jobs.
What my systems run on
Build
- Claude Code (daily driver)
- TypeScript / Node
- Next.js (App Router)
- Postgres (Neon, Supabase)
AI in production
- Claude API · structured outputs
- MCP (servers & clients)
- Agent orchestration
- Prompt-injection hardening
Ship & operate
- Vercel · Railway · VPS
- Stripe billing
- Web push · cron · queues
- PostHog · Umami analytics