Most AI projects stop at the demo
Almost every team I work with has tried AI: prompts in ChatGPT, a Zapier step, maybe a chatbot pilot. Very few have AI doing real work every day, reliably, without someone babysitting it. The gap is rarely the model. It’s the plumbing — clean inputs, clear rules, error handling, logging and someone who owns it.
I build AI automations that run in production: connected to your real tools, with guardrails, monitoring and human approval where it matters.
What I automate
Reporting and alerts
Weekly performance reports assembled automatically from GA4, ad platforms, your store and CRM, with a written summary of what changed. Pacing alerts and anomaly detection so overspend or broken tracking is flagged the same day.
Lead handling
Enrichment, qualification, scoring and routing of inbound leads; call and meeting summaries pushed to the CRM; follow-up drafts ready for a salesperson to review and send.
Content operations
Research, briefs, outlines, first drafts, product descriptions, translations and metadata at scale — always with an editing step before anything is published.
Customer and operations workflows
Support ticket triage, review monitoring, order and inventory alerts, document processing and data sync between systems that don’t talk to each other.
Custom software and internal tools
When off-the-shelf tools don’t fit, I build what’s missing: internal dashboards, admin panels, data pipelines and small web apps.
How I build
1. Map the work
I list the repetitive tasks your team does each week, how long they take and what a mistake would cost. We automate where the time saved is high and the risk is manageable.
2. Ship the smallest useful version
The first workflow typically goes live within two weeks and solves one real problem end to end. Then we expand.
3. Build the guardrails in
Structured outputs, validation, retries and fallbacks. Agents get narrowly defined tools and permissions. Anything customer-facing, financial or irreversible goes through a human approval step.
4. Monitor and maintain
Every run is logged, failures raise alerts, and workflows are updated as your tools, APIs and models change. An automation that fails silently is worse than none.
The stack
n8n is my default orchestration layer — self-hostable, flexible and not priced per task. I use Claude and ChatGPT through their APIs for language work, Python or Node.js for custom logic, and Zapier or Make where a client already relies on them. Everything runs in accounts you own, and every workflow is documented, so you’re never locked in to me or to a single vendor.
Who this is for
- Growth and marketing teams buried in reporting, briefs and manual data work.
- Founders who want AI working inside the business, not just a subscription seat.
- Operations-heavy businesses — ecommerce, agencies, services — with repeatable processes spread across several tools.
Who this is not for
- Anyone looking to replace their team with fully autonomous agents.
- Projects that need model training or research — I build applied systems on top of existing models.