FRENDI MATEO  ·  General Trias City, Cavite · GMT+8 OPERATIONAL

// AI Automation Specialist · E-commerce Operations Manager

AI Automation Specialist who has run the entire operation.

I ran a 70+ person remote e-commerce operation, and I personally built the Python and Claude-powered systems that automated its backend processes. I also built Dominus, an operations app running live for dropshipping businesses.

9+ yrs
Running e-commerce ops
15
Person CS team I built, trained & led
70+
Staff I oversaw
80%+
Disputes won

// The difference

Most developers who can build this have never worked a support queue, fought chargebacks, or managed an entire backend process. I've done all of it.

For over a decade I worked the front line of customer and technical support, then ran e-commerce customer service, fulfillment, and PayPal/Shopify disputes by hand, and managed the whole backend process end to end, long before I automated any of it. That's why what I build fits how a store actually operates, not how a developer imagines it does.

// Selected work

Two production systems for two different businesses. Same approach each time: learn how the operation really runs, automate it step by step, and keep a person in charge of anything a customer or the bank will see.

DTC brand · in production Multi-agent back office One orchestrator and four agents covering fulfillment, fraud screening, disputes, customer notes and a daily P&L.
Dropshipping · in production Dominus An operations system for dropshipping businesses selling into several countries, each with its own language, currency and delivery rules.

// System architecture

Under the hood, that back office runs as one orchestrator and four specialized agents. It runs 24/7, and I've migrated the whole platform across three environments without downtime.

// Data sources

Store ordersplus signed webhooks
Payment processors
Ad-spend data
Bank feed
Carrier tracking

// Orchestration

Schedulertime-based triggers
Safe-run wrappertimeout · retry · lock · audit ledger · alert
Secret resolutionat runtime, no keys in code
LLM agent runtime

// Agents

MonarchOrchestratorrouting · scheduling · budget guard
RajahFinancedaily P&L · ad-spend · billing reconciliation
RegentDisputeschargebacks · bank reconciliation
ViceroyFulfillmentorder tracking · shipping · stale-order & risk audits
ArgusCustomer noteswebhook-triggered personalized order notes

// Outputs + human gate

Human approvalon customer-facing & financial writes
Slackalerts · P&L · ops
Google SheetsP&L · billing
Notionreview queue
Order noteswritten back to store

// Reliability & security, every run

  • HMAC-verified webhooks
  • Runtime secrets, no keys in code
  • Retries + concurrency locks
  • Health checks + daily heartbeat
  • Per-run audit trail
  • Human-in-the-loop
Live · daily operations

DOMINUS

Every order checked. Every customer confirmed. Every product’s real profit, every day.

Cash-on-delivery dropshipping only works if the address is right, the customer knows the order is coming, and someone knows which products actually make money once returns land. Staff were doing all of that by hand across sheets and half a dozen tools. Dominus does it automatically, in every country, and stops for a human whenever it isn’t sure.

Dominus home screen for one business: sales today, and shortcuts for orders, fulfillment, chargebacks, refunds and customer service The same app with the apps menu open: store, helpdesk, bank and ads apps fanned out around the seal button Dominus for a second business: tiles for orders in progress, shipping, delivery outcomes, returns, profit and Ask Dominus The built-in assistant, which answers questions about the app's figures and anything attached
Two live setups, each styled for its business. Figures and names blurred.

// How an order moves

  1. Capture Every order lands on the order sheet the moment it’s placed. Upsells are joined to the order they belong to.
  2. Verify The shipping address is checked against up to three validation sources. Anything unconfirmed is flagged for a person, never guessed. Gate · validated address
  3. Confirm One confirmation per order, in the customer’s language and currency, with the full total. A retry never sends a second message.
  4. Release Orders go to the warehouse only when the address is validated and the stock is there. Out-of-stock orders are tagged, not shipped. Gate · in stock
  5. Measure Profit per product from real outcomes: delivered, returned, still in transit, with product cost, shipping and daily ad spend.

// What the owner gets

Owner app

The whole operation on a phone

Installable app with live sales per store, order status, delivery and return rates, per-product profit, and alerts before a problem costs a day.

Assistant

Ask it, don’t dig for it

Ask by typing, speaking or sending a photo. It answers from the business’s own figures, answers only what was asked, and also answers the team in chat.

Change by describing

Describe it, it gets built

The owner describes a change to the app in plain words. It’s built, tested and deployed, and rolled back on its own if anything breaks.

Supervisor

Fixes first, alerts second

Before the daily warehouse cutoff it re-runs whatever can clear a problem, checks again from scratch, and alerts a person only about what’s left.

Per-country rules

Each market runs on its own

Language, currency, address format and delivery rules are set per country, so a change for one market can’t break another.

Tool-agnostic

Fits the tools you already use

Swapping a tool means rebuilding one connection. So far that’s meant moving support to a new helpdesk, taking a paid store platform out of the order path, and comparing messaging providers across every country.

// Built to be trusted

An upstream tool quietly changed what it sent, and orders stopped qualifying for confirmation without a single error.

I rebuilt the trigger so “not done yet” is decided from the system’s own records, never from a field another tool writes, and added a supervisor that catches that whole class of failure. That’s the standard everywhere in Dominus: every skipped order is logged with a reason, and nothing fails silently.

  • No shipment without a validated address
  • One message per order
  • Human approval on anything unusual
  • Every skip logged with a reason
  • Keys injected at runtime, never in code
  • Private customer data stays private

Already used by a couple of dropshipping businesses selling into European markets. Europe is where Dominus started, not where it stops.

// What I build

Systems I've designed, built and run in production, for a DTC brand and for multi-country dropshipping:

Fulfillment monitor

Catches stuck orders

Flags late or stuck shipments and follows up with suppliers automatically, before the customer has to ask.

Fraud screening

Checks every order

Screens each order for fraud signals before it ships, so the risky ones get held instead of charged back later.

AI-drafted support

Replies that learn

Drafts customer-service replies that get sharper each time a human edits one. The team stays in the loop, the queue gets faster.

Dispute tracker

Never misses a deadline

Tracks PayPal and Shopify disputes end-to-end so a response window never lapses unanswered.

Address check

Nothing ships to a bad address

Every address is validated against several sources before release. Anything unclear is flagged for a person, never guessed.

Order confirmation

One message, every order

Each customer gets one confirmation in their own language and currency, with the full total. A retry never sends a second.

Profit, every day

Real profit per product

A daily P&L posted to Slack, and profit per product from real delivery and return outcomes. No figures pulled by hand.

Owner app

The whole operation on a phone

Live sales, order status and alerts, plus an assistant you can ask in plain words.

// Built not to break

The difference between automation that helps and automation that starts fires: mine wouldn't double-charge or double-email a customer if a step retried, it alerted the moment something failed, and it stayed cheap to run as volume grew.

Built and run in production for a DTC brand and a multi-country dropshipping operation.

// Stack

Claude and Python at the core, wired into the store, payment, support and team tools a business already runs on.

AI / LLM

Claude API (Anthropic) Claude Code ChatGPT OpenAI Grok OpenClaw ElevenLabs BotPress Multi-agent systems

Automation & Dev

Python REST / GraphQL n8n Make.com Web scraping

E-commerce & Payments

PayPal Shopify GraphQL Admin API Stripe Revolut Klarna Wise Airwallex 17Track

CRM, Support & Messaging

WhatsApp Zendesk Freshdesk Gorgias MessageBird Kommo Chatwoot 2Chat WANotifier

Collaboration & Ops

Google Workspace Slack Notion Monday.com ClickUp

// Capabilities

Automation & Dev

Production automation that runs unattended: idempotent, self-monitoring, and cheap to scale; it won't double-charge and it alerts the moment a step fails.

AI / LLM

LLM agents that draft, screen, and decide, and sharpen with every human edit, keeping a person in the loop while the queue speeds up.

E-commerce & Payments

The back office run end to end: orders, fulfillment, fraud screening, and chargebacks won across multiple processors and currencies.

Leadership

Built and led remote teams, from a 15-person CS unit to a 70+ person operation, with hiring, coaching, and process that scales.

CRM & Support

A chaotic support inbox turned into a fast, measured queue: SOPs, QA, and escalation paths that hold up at volume.

Collaboration & Ops

Distributed teams kept aligned: clear docs, tracked work, and the daily reporting leaders actually read.

// Experience

Jan 2017 – Present
Automation Specialist & E-commerce Operations Manager · Freelance
  • Design and build custom Python + Claude automation that runs e-commerce back offices unattended: order monitoring, fraud screening, AI-drafted support, dispute tracking, and daily reporting.
  • Built Dominus, a live operations app for dropshipping businesses: address checks, order confirmations, warehouse release, profit per product, plus an owner app and assistant.
  • Ran the full back office as Operations Manager for a couple of DTC brands, leading the team leaders across customer service, creative, copywriting, ads, fulfillment, and order management: 70+ remote staff. Built and led a 15-person CS team before scaling up to the whole operation.
  • Owned customer service, fulfillment, and PayPal/Shopify dispute handling end-to-end, winning 80%+ of the disputes I represented.
  • Wired the brand's systems into one reliable, self-monitoring operation. It alerted the moment a step failed instead of breaking silently.

Earlier: BPO / customer & technical support · 2006–2017

2016–17
Email Support · TaskUsEmail support for an online store: product info, shipping, payments, refunds; escalated valid issues to the client.
2012–15
Technical Helpdesk Analyst · Xerox Business ServicesIT helpdesk for Xerox employees: access, diagnostics, ticketing, resolution.
2010–11
Subject Matter Expert · Sutherland Global ServicesSupported a team of 9 sales agents: coaching, escalations, call audits, process improvement.
2009–10
Technical Support · TeleTechDSL internet support for North American customers: troubleshooting and configuration.
2007–09
Technical Support · Dell International ServicesPhone/remote support for Dell desktops and portables: repairs, parts, software, peripherals.
2006–07
Technical Support Supervisor · NCO–RMH TeleservicesSupervised order entry and fulfillment; Tier-1 escalations for a gaming-console program.

// Education & certifications

Education

BS Computer Science
AMA Computer College, Parañaque · 1994–1997

Certifications

  • Project Manager: Role & Skills · Metropolitan School of Business & Management UK, 2022
  • Tech Lab Certification · Dell Int'l Services

// Contact

Tell me where the manual work is. I'll show you what I'd automate first.

AVAILABLE FOR NEW WORK ·General Trias City, Cavite, Philippines ·GMT+8 · works comfortably on US hours