Work
Examples of systems I've built and shipped. Real clients, before and after.
Invoice Processing Automation
The context
A business receiving 20 to 30 invoices a week by email. Every one was handled the same way: someone opened the email, opened the PDF, retyped the numbers into a spreadsheet, then flagged it for approval. Same four steps, every invoice, every week.
What I built
When a new invoice hits Gmail, the workflow picks it up automatically. Claude Vision reads the PDF (or PNG, or JPEG) and pulls out the vendor, amount, line items, and due date. That data goes straight into Airtable. If the amount is over €5,000, a Slack button fires for approval before anything gets marked as processed. Duplicates are caught before they create a second record.
Result
The finance team went from 3 to 4 hours of manual work every week to a 15-minute check, with invoices coming in the same way and nobody having to touch them.
Stack
n8n, Claude Haiku (Vision), Airtable, Slack, Gmail
Multilingual Content Pipeline
The context
A media company was taking US YouTube videos and turning each one into French and Spanish content across three formats: a script, a social thread, and a book chapter. Someone was doing all of this manually, for every video that came out.
What I built
An RSS feed watches the target YouTube channels and triggers the workflow when a new video goes up. Supadata pulls the transcript. Claude Sonnet rewrites it in both French and Spanish, producing all three formats in a single pass. Everything lands in Airtable for the editorial team to review and publish. The whole cycle runs on its own every two hours.
Result
What used to take a full day of work per video now runs without anyone touching it. The team reviews and publishes instead of translating and reformatting.
Stack
n8n, Claude Sonnet, Airtable, Supadata
Email Urgency Monitor
The context
The inbox had become unworkable. Newsletters, platform notifications, and automated emails were piling up alongside things that actually needed a reply. The only way to know what mattered was to read all of it.
What I built
n8n checks Gmail every 15 minutes. Claude Haiku reads each new email against a custom urgency rubric and classifies it. Anything urgent gets a draft reply written and staged in Gmail, ready to send with one click. A Slack digest summarizes everything that came in. A blocklist of 50+ domains filters the obvious noise before it even reaches the classifier.
Result
The system identifies what needs attention and surfaces it automatically. Inbox review time went from 30 minutes a day to under five, and nothing urgent slips through.
Stack
n8n, Claude Haiku, Gmail, Slack
Ready to automate?
Book a 30-minute call. We'll map out your workflow and leave with a clear picture of what to build and how long it takes.
Book a call