AVAILABLE FOR AUTOMATION & INTEGRATION ROLES
I build the integration the software doesn't offer.
I uncover the backend requests hidden behind business software and turn them into stable automation services. When official integrations stop at the surface, I inspect the layer underneath.
AI Automation EngineerProduction backend automationNode.js + PlaywrightAPIs + n8n + ZapierPhilippines → US Remote
Method — layer descent
When the surface fails, don't go around the problem — go underneath it, one layer at a time, until something solid holds.
L1 · Surface — workflow tools
Start with the supported integration.
✕ the action exists — the field the business needs isn't exposed
L2 · Interface — browser automation
When the integration stops, inspect the interface.
reaches the data — until a selector or a session changes
L3 · Traffic — network inspection
The interface is only one layer above the request.
one authenticated request returns everything the UI was scraping
L4 · Foundation — backend service
Build the reliable path when the supported one doesn't exist.
✓ in production — reporting runs on schedule
Method — layer descent
Selected systems
Three systems, three levels of maturity — labeled honestly. The flagship runs a US company's daily reporting.
Investigated the authenticated application requests used by a field-service platform and built a stable internal reporting service around the company's authorized access.
Production reporting system used daily by a US HVAC operations team.
Unstructured WhatsApp messages turned into structured Monday.com records — with a human approving every write.
An AI concierge answering guest questions from the hotel's own documents.
The owner and operations team needed daily visibility into installs, sales, labor costs, and job performance. The data lived inside their field-service platform — and the reports they needed didn't exist in it.
Started with workflow automation against the official API — the data wasn't there. Moved to Playwright browser automation: it reached the data, but UI selectors and session handling made it fragile. Inspecting the browser's network activity during the company's authorized session identified a more reliable authenticated export and request flow — the same requests the web application itself uses. That became the foundation.
Manual reporting and fragile UI automation
Scheduled backend service feeding daily operations dashboards
Operational data arrived as free-form WhatsApp messages, requiring manual reading and re-typing into Monday.com.
Confidence scoring so only low-certainty extractions require review — approving everything doesn't scale.
Guests ask the same questions around the clock — policies, amenities, local recommendations — and every answer consumed staff time.
Public repo on GitHub.
Structured retrieval (chunking + embeddings), tests around the services layer, and proper session handling for the admin view.
Workflows
Demo builds — self-directed pattern explorations, not client production systems. Included to show how I reason about data flow, joins, branching, enrichment, loops, and failure handling.
Self-directed demonstration
A two-source reconciliation pattern: fetch inventory and sales, join on product ID, and branch to alerting or a stock-update path. Built to explore data-integrity checking between two systems.
Self-directed demonstration
A warehouse audit pattern: pull two warehouse feeds, join them, reconcile the differences in a code step, and write an audit trail to a sheet.
Self-directed demonstration
An order-to-delivery orchestration pattern: check stock, raise a purchase order, refresh inventory, then merge and notify — with a conditional path on the stock check.
Self-directed demonstration
A modular workflow that accepts a blog URL, structures the article with an AI model, generates title and section visual assets, stores the files in Google Drive, and passes normalized metadata into a separate Google Sheets storage workflow. Identifiers and endpoints sanitized.
Method
The descent above is the method. Day to day, it reduces to three rules.
Native integrations and public APIs when they expose the data. No descent for its own sake.
When UI automation drifts, the failure shows where the software actually moves its data — one layer down.
Wrap what holds in a deployed service with error handling, monitoring, and a clear owner.
I use AI throughout planning, implementation, debugging, documentation, and review. I remain responsible for understanding the system, validating assumptions, testing the output, and owning what reaches production. AI multiplies the engineering — it doesn't replace the judgment.
Core capabilities
About
I'm Ed — based in Iligan City, Philippines, working with US businesses. Before engineering I worked in customer service, telesales, and business development, which is why my technical updates make sense to business owners, not just engineers.
The work I care about sits in one place: operations-heavy businesses running on industry-specific platforms whose most important data is locked behind incomplete or missing APIs. Getting that data out reliably is the problem I keep coming back to.
I moved from workflow automation into deeper backend automation after hitting real API limitations — and discovered that unfamiliar, poorly documented problems are the ones I enjoy most.
Off the clock: the gym, good conversations, and an unreasonable loyalty to graham mango float.