Zacao sells chocolate through DTC, Amazon, wholesale, and B2B. I run growth operations there, and I've replaced the most repetitive parts of that job with five agents that call retailers, chase invoices, send outbound, and watch our orders. Together they're the automation layer of Zacao's revenue engine, and I built it end to end.
"I build production AI agents" is an easy claim. Below is each system, what it replaced, and how it runs.
Our team used to qualify DTC brands and retail partners with a phone list and a spreadsheet. The agent makes those calls now, asks the qualification questions, and writes the answers into the HubSpot fields the team already reads.
account list → Retell AI places the call (ElevenLabs voice) → qualification answers extracted → written to HubSpot properties → flagged for a human when a call needs judgment
Issues invoices, checks each against its PO dates before finance sees it, and chases the late ones. The recurring billing discrepancies stopped.
Shopify order data → invoice issued → dates verified against the PO → mismatches flagged → follow-ups on schedule → reconciled in Finaloop, orchestrated in n8n
Finds contacts, verifies addresses, writes each email individually, and sends on schedule while I do other work. Replies come from founders and department heads.
Clay sources and enriches → emails verified → one message per contact → sent via Gmail API on schedule → outcomes logged in HubSpot
Watches the order-to-ship workflow across four channels and raises the exception before a customer feels it.
order status across DTC, Amazon, wholesale, B2B → exception rules in Zapier and n8n → alert raised early → recurring failures get a documented SOP
Built for my own search: sources roles daily, scores each on skills, comp, location, and visa safety, drafts the outreach. Other job seekers now pay for it.
postings collected daily → scored against the rubric by Claude → shortlist with reasoning → outreach drafted per role → orchestrated in n8n
A few months into Zacao, I found 42% of our SKUs blocked from fulfillment by unresolved stock flags. No one had asked me to look.
Product that should have been moving sat still, and the flags never appeared in any report leadership read. I cleared the backlog, then rebuilt the order-to-ship workflow with documented SOPs and automated alerts so the same failure can't recur quietly. That rebuild became the alerting layer above.
Every agent on this page started the same way: a problem I found in the data before anyone assigned it to me.
Built around live coaching instead of static content. I wrote the app, the backend, the Cloud Functions, the payment integration, and the push infrastructure.
Three audiences share one codebase, and each gets its own path through login. Members subscribe and attend daily classes. Coaches are subscribers too, with their own onboarding, their own collection, and an earnings screen that breaks down what they made per referred user. Admins run the whole thing from inside the app, including sending push without ever opening a console.
That made the login gate a real state machine: authenticated, then which role, then subscribed or lapsed, then onboarding complete or not, then expiry still valid. A lapsed coach gets bounced to payment the same way a lapsed member does. Most of my hardest work went into that routing rather than into any screen.
My first real product. Course information, events, clubs, and announcements in one place, instead of the four channels students were already checking.
Same serverless shape I'd reuse four years later, learned the hard way here: Firestore for the data, Firebase Storage for club and event media, Cloud Functions for the logic that couldn't live on the phone, FCM for announcements students would otherwise miss, Google Maps for event locations, and REST calls out to the systems that held the rest.
Most days that's Clay, HubSpot, Zapier, and n8n for the GTM plumbing, with Claude and Codex to build the agents themselves on Retell AI, ElevenLabs, and LLM APIs.
The data work happens in SQL and Python, with Snowflake, Metabase, Looker, Power BI, Tableau, and Google Analytics depending on who's reading the output. The product side is Flutter, Firebase, and Cloud Functions; the commerce side runs through Shopify, Amazon FBA, Finaloop, Notion, and Jira.
The fastest way to evaluate me is live: the n8n canvas, the Retell dashboard, the Clay table. I'm in Brooklyn, available from May 2026, looking at GTM engineering, revenue operations, and growth roles.