Alex Garcia
Practical résumé · prepared for LIXIL · remote from Calgary

From the job site to the screen: AI & digital operations

For 15+ years I built and renovated homes — the rooms your products live in. Since January 2026 I have used AI agents to build websites, apps, video and automations, with a person approving everything before it goes out. This page shows the work itself, with links.

See the concept I imagined for American Standard → Watch the short video One-page CV

Why LIXIL

“I see AI as a ‘power suit’ because it can strengthen individual capability … This is not about reducing people.” LIXIL CEO, Integrated Report 2026, p. 11 — which also describes the move to the “democratization of AI” (p. 53). Digital strategy · read 2 Nov 2026

That is how I already work: people who are not engineers using AI to build real tools — safely, with clear rules, human approval and costs agreed before they are spent. I also know the other end of your products: the job site, the installer, the contractor and the homeowner choosing a finish. I would love to help LIXIL connect the two, remotely from Calgary.

Something I imagined for you

Not a critique — an idea, built so you can click it instead of reading about it.

Concept prototype: a calm bathroom planned room by room
Concept prototype · not commissioned · not affiliated with LIXIL

Plan your bathroom, room by room

A planning experience imagined for americanstandard.ca:

  • an opening where the lights come on and the room fills as you scroll;
  • real homeowner questions answered from American Standard's own product pages, each with its source;
  • a planner that suggests a suite of real products and a spec card to share with the installer;
  • a view for the trades, from someone who spent 15 years on job sites.

In the prototype the answers are pre-written from public product information. A live version would use an AI model grounded in the catalogue, always citing its source.

Open the concept →

The work in 85 seconds

Real screens from real projects. The opening and closing scenes are AI-generated and labelled as such.

Download the full-HD version

Where I could help

Three remote roles where my work and my trade background meet. Open to the shape that fits your team.

Digital & marketing operations

Product pages, campaigns and short video produced faster with AI — with review and approval built in before anything is published.

Proof: bilingual campaign end to end; websites; AI video with a cost guard.

Sales enablement & trade content

Installation explainers, showroom and contractor material, product Q&A with sources — in plain language the trades trust.

Proof: 15+ years on job sites; trailers and reels; the concept above.

AI enablement for teams

Helping non-technical colleagues turn routine work into safe, documented AI workflows: human approval, cost limits, independent checks.

Proof: the agent operating system and the approval-first robots below.

Proof of work

Each card says what it really is: own product, client site, volunteer project or internal tool. No invented clients or numbers.

Internal · in daily use

An operating system for a team of AI agents

Problem
Several AI coding sessions working on one live website at once will overwrite each other and ship mistakes.
What I built
Shared rules and a start-of-session briefing, checks that block publishing until an independent review passes, shared state across parallel workspaces, reusable skills and specialist sub-agents.
Outcome
One person runs many agents safely on a production repository: 3,600+ commits since April 2026.
  • Claude Code
  • Codex
  • Git worktrees
  • Hooks
  • Linear
Try the approval pattern (sample data)
Own product · private beta

Portage — a job-search copilot

Problem
Job seekers lose hours finding the right postings and starting each application from a blank page.
What I built
A web app that reads official employer job feeds, ranks postings for one person's profile, and drafts a CV, cover letter and interview prep per role for the person to review. Built first for one real job seeker.
Outcome
One recorded run read 2,291 postings from 51 feeds. Per-user monthly AI budget built in.
  • Cloudflare Workers
  • D1
  • Claude API
  • Tool use

Honest status: AI drafting is paused while API credit is renewed; manual materials work.

autonomadigital.com/portage
Own product · pre-launch beta

Unhurry Home — a calm family app

Problem
Family apps ask for bank passwords and send constant notifications; the money, the week and the meals still live in one parent's head.
What I built
A private family home base: join with a family code and PIN (no email, no password), personal planning stays on the device, plus four free tools that run in the browser — Subscription Sweep, Sunday Reset, Life in Hours, Dinner Plan.
Outcome
Live; tested end to end in July (create family → sign in → delete). Not launched publicly yet.
  • Cloudflare Workers
  • D1
  • PWA
  • Local-first
unhurryyourself.comhome.unhurryyourself.comSweep
Client site (live) + concept studies

Websites that feel like the place itself

Problem
Small businesses do beautiful work that deserves to be seen the way their customers experience it.
What I built
A live site for a family photographer built around her own archive (black-and-white with a colour switch, a short film from her photos, English and two kinds of Portuguese), seven concept studies across sectors, and a lab of interaction effects.
Outcome
Real, clickable work instead of screenshots. Two more sites for friends and family are shown on request.
  • Claude Code
  • GSAP
  • Cloudflare Pages
  • AI imagery
vivianedias.comautonomadigital.com/workLabs
Volunteer project · delivered

A bilingual campaign, end to end

Problem
A Toronto church had one week of events in October and no media team.
What I built
Event page and a six-page team portal, trailer and teasers, a seven-night countdown in Portuguese and English, posters, print, an ads plan checked before any spend — then a team portal for a second series with one approved video a day.
Outcome
Delivered across the event weeks in both languages, with the church approving each piece.
  • ffmpeg
  • Python
  • Cloudflare Pages
  • Meta Business Suite
Event pageTeam portal
Internal automations · running

Robots that do the routine, people who approve

Problem
A one-person studio can't check two inboxes, social media and site visits by hand every day.
What I built
Twice-daily email triage that only drafts replies (never sends), an Instagram engagement robot that follows a weekly plan the owner approves, a weekly site-visits report, a "watcher" that checks each robot really produced its result, and a research app that transcribes videos and answers with links to the exact second.
Outcome
Routine checks happen on schedule; the owner only reviews what matters.
  • Python
  • launchd
  • Chrome DevTools Protocol
  • Whisper
  • Claude

Private data, so no public link. Sample versions with fictional data are on GitHub.

GitHub portfolio
Internal pipeline · in use

AI video with a cost guard

Problem
Generative video is easy to overspend on and easy to ship with visible defects.
What I built
A Google Veo / Imagen pipeline that quotes the cost before each generation, reserves budget, allows one take per image and reviews frames before anything is used; plus a video assembly toolkit for trailers and reels.
Outcome
A 12-clip batch for under US$10; one defective clip caught in review before use.
  • Vertex AI · Veo 3.1
  • Imagen
  • ffmpeg
  • HyperFrames
See the campaign trailer

How I work with AI agents

The agents draft and build. I set the goal, the rules and the checks — and I answer for the result.

  1. Start from the finish lineWrite down what "done" looks like and how we'll check it, before any agent starts.
  2. Right agent for the jobClaude Code for building, Codex for a second pair of eyes and audits, Grok for quick outside research — run in parallel when the work is independent.
  3. The builder never checks its own workAn independent reviewer checks the real page or app, on mobile and on desktop, before anything ships.
  4. A person says yes before anything goes outNo email, post or customer message is sent by a robot without approval. Costs are quoted before money is spent.
  5. Honest labelsPrototype, beta, live — always said plainly. Handoff notes so the next person (or agent) never starts from zero.

Skills

AI & agents

  • Claude Code · Codex · Grok
  • Claude API: tool use, prompts, cost limits
  • Retrieval with sources (RAG)
  • Google Vertex AI · Veo · Imagen
  • Local transcription (Whisper)

Automation & web

  • Python · JavaScript · Bash
  • Cloudflare Workers · D1 · Pages · KV · Vectorize
  • Browser automation, REST/GraphQL APIs
  • Scheduled jobs, approval workflows
  • Git · Linear

Content & people

  • Websites in English and Portuguese
  • Trailers, reels, print and signage
  • Led crews of up to 10
  • Budgets, vendors, subcontractors
  • Plain-language handoffs

Path

Jan 2026 – nowFounder, Autonoma Digital · Calgary — small AI studio; part-time from January, full-time since October 2026.
2024 – 2026Carpenter, Gunthers Building Centre · Calgary
2010 – 2021Co-founder & Project Manager, Real X-Pertise Investment · Toronto — renovation projects, budgets, vendors and trades.
2005 – 2021Finish Carpenter & Foreman, Penegal Trim & Supply, DCC Carpentry · Toronto — led up to 10 carpenters.
EducationAnhembi Morumbi University — Business Administration (incomplete, 2004) · First Aid · WHMIS