Guskage.quest

AI & Automation Engineer

Hours into minutes. Prototypes into production.

I take the repetitive work that quietly eats a team's week and get rid of it. When an automation platform can't take it further, I build the software that can.

09Engagements
08Sectors
03Past the ceiling

Remote — Philippines, UTC+8 · AU/SG hours, US overlap to midday

Also open to AI and automation engineering roles. CV

Illustration of Gus holding Kage, a black pug
Kage is online

How I work

Three shapes of work

Not a sequence — each is a different answer to what a problem needs.

01

Automate

Work that repeats, done by a person, on a schedule nobody enjoys. It runs without anyone touching it now.

02

Integrate

Tools and people that hold the same information and never talk. They talk now.

03

Build

The point where a platform stops being enough. These needed software written.

Automation first, because it is faster and cheaper for you — and custom software only when the platform hits its ceiling.

Case studies — anonymised

Nine engagements, one page

Left is what it was. Right is what it is. No client names, no logos, no invented numbers.

Automate

03

Work that repeats, done by a person, on a schedule nobody enjoys. It runs without anyone touching it now.

Insurance broker

1 mo

Carrier documents file themselves into the CRM.

OpenAI · Zapier · Google Cloud

Agriculture research

1 mo

Trial plots compared automatically; the difference comes out as data.

Python · OpenCV · NumPy

Agriculture

2 mo

Live satellite weather sets the spray and irrigation schedule.

Tomorrow.io · Google Cloud · Looker Studio

Integrate

03

Tools and people that hold the same information and never talk. They talk now.

Law, interior design, accounting

Ongoing

One new client creates the matter, the tasks and the invoice.

Zapier · Make · HubSpot · Zoho · Clio · ClickUp

White-label SaaS

2 mo

Calls answered and booked at any hour, logged either way.

VAPI · Retell · Make · Twilio · HubSpot

Medical clinic

1 mo

A paper practice replaced end to end.

AppSheet · Apps Script · Looker Studio

Build — past the ceiling

03

The point where a platform stops being enough. These needed software written.

Software team

2 mo

Every pull request arrives already summarised.

No platform reads a diff against a whole codebase.

Python · Bitbucket API · Webhooks · OpenAI

Web security

3 mo

Transactions scored as they happen, flagged on the spot.

Scoring sits on the payment path itself.

C#/.NET · MaxMind API

Website building

6 mo

A CRM record becomes a finished site in minutes.

Started on the platform. Outgrew it.

Python · OpenAI · Ideogram · Make · Bitrix24 · Duda

03 of 09 needed software written · ask Kage about any row

Own work

When automation isn't enough, I write the software.

Four projects with no client and no platform ceiling — this is what happens after the no-code tools run out.

Offline-first budgeting app

Deployed, signups closed

Envelope budgeting built local-first: every screen renders from the device, every write lands there first and replays to Postgres when there's signal. Reports run off the same local ledger, so they work with no signal too.

Money software has to be right when the network isn't. Amounts are integer cents everywhere, and the ledger rules are enforced twice — once in TypeScript, once in the database with deferred constraint triggers — because one of those will eventually be wrong.

Next.jsTypeScriptSupabaseRechartsPlaywright
75 files, 10,166 lines, 9 tables
  1. Envelopes screen: unallocated money at the top, then Bills envelopes with balances and progress bars against their monthly targets.

    01Every peso has a job

    Payday fills the envelopes. What is left over sits in Unallocated until it is given one.

  2. Add transaction screen with 1250 typed as the amount, SM Supermarket as the payee, paid from BPI Checking and spent from Groceries.

    02Spend from the envelope

    A grocery run, logged from the Groceries row. The envelope and account are already chosen.

  3. Envelopes screen after saving with no signal: an amber banner reads Offline, everything still works, 1 change waiting, and a toast confirms 1,250 pesos spent from Groceries.

    03No signal. Saved anyway.

    The write lands on the device and the balances move. The banner says one change is waiting for the server.

  4. Reports screen for August: spent, income, net and budget used tiles, then income versus spending bars for six months.

    04Where it went

    Last month by envelope and group, against the budget, with a pace estimate while the month is still running.

Config-driven event sites

Ran a real event

One codebase, one config file per event: RSVP, a gift registry with claims and fund pledges, a photo walkthrough to the venue, and an admin side for whoever is running the party. Twenty swappable themes sit behind a written contract; the celestial one ran a real first birthday.

It was built for one party, then generalised. An event is a theme plus an event type plus a single config file; a feature that is switched off has no route at all, not a hidden one; and seeded rows carry a test flag so a rehearsal never counts as a guest. It is now being rebuilt as a multi-tenant product for event coordinators.

Next.jsTypeScriptPostgresPrismaTailwind
20 themes, 16 routes, 5 tables
  1. Invitation home screen in the celestial theme: a baby's portrait in a gold ring under the name Isla Estelle Marin, a three-line poem, and the welcome message.

    01The invitation

    What a guest opens from the chat link. Twenty themes swap behind this page without touching it.

  2. RSVP form filled in: Maria Santos, attending, 2 adults and 1 child, with a message in Taglish.

    02RSVP in twenty seconds

    Name, yes or no, how many adults and children, a note. Nothing to sign up for.

  3. Gift registry with illustrated cards for board books and storybooks, one marked all claimed, the other 2 of 3 claimed.

    03Gifts, without duplicates

    Guests claim from the registry and a gift that is full says so. The education fund takes pledges.

  4. Admin RSVP list on a phone: 31 attending, 8 not attending, 39 total, a Download CSV link, and guest cards with their messages.

    04Headcount for the organizer

    The admin side on a phone: who is coming, how many, every message, and a CSV to hand the caterer.

Short-form video pipeline

Self-hosted

A dashboard and a scheduled dispatcher driving six agents into rendered videos.

This is the whole argument in one project. I ran it on Airtable and n8n first, hit the limits those tools have, and replaced my own setup with software. I am not selling you away from platforms I don't use — I use them until they stop paying off.

6 agents, SQLite + Drizzle

RemotionNext.jsDrizzleClaude API

KageKit

In use

The harness the rest of this list is built with: one repo that sets up a machine and starts a project, both reading the same agent definitions so the two cannot drift apart.

AI assistance is only worth having if it is governed. Model routing is enforced rather than hoped for: cheap models do the reading and the mechanical edits, expensive ones only plan and review, and every agent shows its tier while it runs, so the routing is checkable instead of aspirational. Each session gets its own worktree, which is what keeps a half-finished experiment from ever landing in the wrong branch.

Two layers, one shared routing table

Claude CodePowerShellGit worktreesGitHub template

About

The technical person inside teams that weren't technical

The automation is the easy part. Working out what the business actually does is the hard part.

Illustrated portrait of Gus

Computer Science background, and then the technical person inside teams that weren't technical — a national LiDAR programme, a science high school classroom, an agribusiness field operation, and client systems since. In every one of them I was automating my own work to get it done. Full time on automation and software since 2020.

That's the useful part. The automation is not the hard bit; working out what the business actually does, as opposed to what the process document claims, is the hard bit. I've had to do that in four industries, each time as the person who then had to build the thing.

Career

2012 — now

Independent Engineer and Automation Consultant — Part time until 2020, full time since

Client systems across law, insurance, medical, security, agriculture and SaaS. Ran alongside the roles below, not after them.

2021 — 2024

AgriTech Coordinator — Kennemer Foods International

Led teams applying technology in a working agribusiness. Field realities first, software second.

2018 — 2019

Chief Science Research Specialist — UP Mindanao GeoSAFER 5

Led a research team and delivered every objective inside the defined timeframe.

2017 — 2018

Special Science Teacher — Philippine Science High School

Taught Computer Science and Research to senior-year students.

2014 — 2017

Senior Science Research Specialist — UP Mindanao Phil-LiDAR 1

Managed the Data Processing Component of a national LiDAR programme.

2007 — 2012

BS Computer Science — University of the Philippines Mindanao

PHILNITS (Japanese IT standard exam) · JLPT N4.

Skills

AI & agent systems

Claude APIClaude CodeOpenAI CodexModel Context Protocol (MCP)Agentic tool-use loopsRAGEvaluation harnessesPrompt engineeringOpenAI APIVAPI / Retell (voice)

Automation platforms

n8nMakeZapierGoogle Apps ScriptAppSheet

CRM & business systems

HubSpotZohoBitrix24Monday.comClickUpNotionClio

Languages

TypeScriptPythonJavaScriptC# / .NETSQL

Data & backend

PostgreSQL (RLS, migrations, pgTAP)SupabaseCloudflare Workers & D1SQLiteIndexedDBREST and webhook integration

Web & app

Next.jsReactTailwindOffline-first PWAsService workers

Quality & delivery

VitestPlaywrightpgTAPGitHub Actions CISpec-driven deliveryCode review

Leadership & enablement

Team leadership (delivery, cross-functional and research teams)Spec-driven delivery with stakeholder sign-offCode review and mentoringTraining non-technical usersTechnical writing and documentation

Cloud & infrastructure

CloudflareVercelGoogle Cloud PlatformDocker

Geospatial — Available on request

ArcGISQGISGoogle Earth EngineLiDAR processingOpenCV

Also open to AI and automation engineering roles. CV

Engagement

How this goes

Three stages, plain terms, and a call if it fits.

  1. Discovery week

    Paid, and short. I sit with the people doing the work and map what actually happens, which is never quite what the process document says. You end the week with a written scope, a recommendation, and an honest read on whether this is worth building at all.

  2. Build

    Smallest useful thing into production first, so being wrong is still cheap. Automation where a platform does the job; code where it doesn't. You see progress weekly, not at the end.

  3. Handover

    Documentation written for the person who inherits it, and a walkthrough with whoever will run it. The work has to survive me leaving, or it wasn't finished.

1–6 months

Typical engagement length. Some run longer as a retainer.

Fixed scope or retainer

Fixed for a defined build, retainer for ongoing work.

Remote, UTC+8

AU/SG hours, with US overlap to midday.

How I build

The system cannot produce wrong data.

The rules that matter are enforced by the database itself and not just by the screen, so a mistake in the interface still cannot corrupt your records.

New work does not break old work.

Every change has to pass an automated check before it ships, so you are not paying me later to repair something that already worked.

You approve the plan before I build it.

Each stage is specced and signed off first, so you find out what you are getting at the start rather than at delivery.

One person, the output of a small team.

I direct AI coding agents and review everything they produce, so more gets built per week without adding headcount.

What if the discovery week says not to build it?
Then that's the deliverable, and it's the cheapest good news you'll get all year. It has happened. Some problems are a process change and a spreadsheet, and you should hear that from someone before they invoice you for six months.
Do you only work with the tools we already have?
I start with them. Most businesses already own more capability than they use, and the fastest win is usually connecting what's paid for. I'll tell you when the tool is the thing standing in your way.
Who owns what you build?
You do — code, accounts, credentials, all of it in your name from the start. No part of it is hosted somewhere only I can reach.
How do we work across time zones?
I'm UTC+8, so I work AU and Singapore hours with US overlap to midday. Most work is asynchronous with a standing weekly call; I'll flex the call to your hours, not mine.
Book a call

Contact

Tell me what your team keeps doing by hand.

Describe the work and I'll tell you whether it's worth automating, worth building, or worth leaving alone. Replies within 24 hours.

dev@kage.quest

Messages sent through the form or the chat are processed by Anthropic and Resend to deliver them. Nothing is used for anything else.

Kage

Kage

Gus's loyal pug

*woof woof!* Hi! I'm Kage 🐾

Gus's pug and hype dog. Ask me anything about my best friend — I'm very motivated to get him hired. Treats depend on it.