Ryan L.

Most broken automations aren't tool problems. They're systems problems: inputs that don't match, duplicate records, logic nobody wrote down, and failures nobody sees. Closing that gap is the whole job. What I've built: - A two-way QuickBooks ↔ Jobber sync. A matching table means retries never create duplicates, and a scheduled reconcile check flags anything that doesn't line up. - An AI quoting system for a contractor. Staff describe the job in chat, a four-step pricing engine fills in the numbers (their own price list first, and it flags anything it can't price), and a draft quote lands in Jobber for review. - A voice-first AI assistant for a business owner. It reads and drafts email behind an approval gate (it can't send), previews calendar changes before saving them, answers questions from company docs and spreadsheets, and delivers a morning brief built from daily reports. - AI email triage across several shared inboxes: every email classified, duplicates merged, attachments read and filed by customer. - Fixed-price audits of fragile automations and Apps Script, with severity-ranked findings before anyone pays for a rebuild. How I build: - AI is one validated step, not the whole system: structured output, checked by code, with a person approving anything customer-facing. - Every write is safe to retry, and failures are logged and alerted instead of failing silently. - Least-privilege access, on accounts and servers you own. At handoff, nothing depends on my accounts. - Plain-English handoff notes and recorded walkthroughs. Stack: n8n, Zapier, Python, JavaScript, Google Apps Script, Claude & OpenAI APIs, Postgres, Google Workspace, Microsoft Graph, HubSpot, Jobber, QuickBooks, Shopify. Send me the messy version: the formats that don't match, the process held together by hand. That's where I do my best work.