devbeans

Great coffee takes the right beans. Great software takes the right developer.

AI engineer · senior software engineer & architect

Albert Maranian Jr

Claude Code · Laravel · React / Next.js

I orchestrate AI to ship complex business software ten times faster, with the architecture, tests and review that twenty years of writing code by hand taught me.

AI-orchestrated delivery · Senior architecture · Tested end to end

ERP · SaaS · Accounting · Payroll · Inventory · Business operations

Building production software since 2006 · Davao City, Philippines · remote across Asia Pacific

How I build now

AI engineer, not a vibe coder

I've written code for twenty years and I've enjoyed every year of it. I still own the parts that matter: what gets built, how it's structured, and whether it's good enough to merge. What changed is who does the typing. I now orchestrate Claude Code agents to write the feature, its unit and feature tests, and the end-to-end browser tests. Each works in its own lane, and a second agent checks the work before it reaches me. Shipping has always been my greatest asset, and with Claude I ship features and whole products roughly ten times faster than I did typing every line myself. I'm enjoying it more than ever.

Vibe coding How I work
Prompt until it looks right The ticket is the contract. Every requirement is traced to evidence.
Trust the output One agent writes, another verifies. Nothing self-approves.
Tests if there is time Unit, feature and end-to-end tests ship in the same pass as the code.
Context lives in the chat Team standards live in skills and an MCP server the agents query.
Can't explain the diff Twenty years of reading code. I own the architecture and every merge.

The loop, on every ticket

  1. 01

    Pick up

    Linear MCP

    The agent ranks the board and I choose the ticket. It moves the status and cuts the branch. From here on the ticket is the contract: every claim of "done" has to trace back to a sentence in it.

  2. 02

    Plan

    planning agent

    A planning agent drafts the approach against the ticket's own text: schema, services, edge cases, test plan. Nothing gets built until I approve it. The plan is where twenty years of architecture judgement does its work.

  3. 03

    Build + test

    implementation agent

    An implementation agent writes the feature along with its unit and feature tests (PHPUnit / Pest, Vitest), following team conventions it loads as skills rather than guesses at.

  4. 04

    Verify

    verifier agent

    A separate verifier agent builds a ledger that marks each requirement as met or not met, with evidence a sceptic could check. The agent that wrote the code never approves it.

  5. 05

    End-to-end

    Playwright · Chrome DevTools MCP

    The agent drives the real app in a real browser (clicks, forms, console and network) and confirms the feature works for the person who asked for it.

  6. 06

    Review + merge

    /larament-review

    My review skill diffs the PR and runs Laravel and Filament review lenses in parallel. It verifies every finding, posts the ones that hold up as inline GitHub comments, and moves the Linear ticket back to Todo or on to Merge Ready. I make the merge call.

The toolkit

Most-used skills

from my own session logs
  • /larament-review
    PR review and Linear board sweep (my own)
  • /code-review
    parallel finders, then a vote to verify each finding
  • /ui-ux-pro-max
    design intelligence for every interface

Skills I wrote

workflows, codified
  • larament-review
    Laravel + Filament review lenses; tickets that bounce twice with only low-severity nits get waved through
  • linear-next
    ticket → plan → build → verify → browser test, with the issue as the contract
  • backend-laravel
    a team's Laravel conventions as agent instructions: Snowflake IDs, FluentRule, tenant scoping, Pest
  • frontend-nextjs
    the matching Next.js conventions: ApiRequest builder, SWR hooks, form field wrappers
  • resolving-judgement-call-tickets
    picking between competing implementations, with reasoning a lead can read cold

MCP servers in the loop

the agents' hands
  • Standards MCP
    a Laravel-native MCP server I built that lets any agent query a nine-person team's coding standards
  • Linear · GitHub
    tickets, status, PRs, inline review
  • Laravel Boost
    version-matched docs, schema, logs, tinker
  • Playwright · Chrome DevTools
    end-to-end runs and live debugging
  • Context7
    current library docs instead of stale memory

Case studies

four systems · four industries

Twenty years building the software companies actually run on — ERPs, SaaS platforms, and line-of-business systems carrying real financial and compliance weight. Four of them below, in four industries: food service, real-estate brokerage, real-estate development and construction, and pharmaceutical distribution. Each built end to end, from schema and domain logic to the interfaces people work in every day. All still running.

01

AJA Savory Food Services

Multi-Entity ERP · food service

Three businesses — subscription meal delivery, multi-branch retail with POS, shared corporate back-office — on one Laravel + Filament panel. AI marketing pipeline, Philippine payroll, event-driven accounting, Loyverse integration.

sole engineer ~97k LOC · 142 models 1,358 commits · 2.5 years still active
Read the case study →
02

Marrea

Real-Estate & Construction ERP

The full lifecycle of a housing project: sales, decade-long receivables, treasury, procurement, construction costing, HRIS and statutory payroll — across two Filament panels. HRIS + payroll owned end to end, plus review authority across a five-developer team.

senior engineer & tech lead ~240k LOC · 361 models team of five · 1,203 of 7,671 commits mine two panels
Read the case study →
03

Comtrak + Cubed

Real-Estate Brokerage Platform

Full-stack ownership of both tiers: a Laravel API with two Filament panels and an automated commission engine that computes VAT-stripped, tax-deducted splits the moment a sale is recorded — plus a Next.js 16 agent app with its own design system, and an AI agent whose authorization lives in its role-scoped tool surface.

sole engineer, both tiers ~42k LOC backend · 104 components frontend 308 commits 21 AI tools · 4 personas
Read the case study →
04

Priomed

Pharmaceutical inventory

Multi-branch stock, procurement, controlled adjustments, and dispensing with a pharmaceutical-grade audit trail — deliberately compact, with layered services, global-scope branch isolation, and a static-analysis ratchet.

sole engineer ~14k LOC · 33 models 150 of 154 commits the architecture case study
Read the case study →

Experience

20+ years · 2006 – present
2023 – now

Independent Senior Software Engineer

Freelancer · Davao City, Philippines

Design and delivery of custom business systems, SaaS platforms, and ERP solutions, from requirements through deployment and support. Implementation, testing and review are orchestrated through Claude Code agents. Clients in real estate, food services, retail, and pharmaceutical distribution. Leading contributor on contract to a nine-person workforce-management SaaS (Laravel 13 + Next.js 15), where I built the team's Standards MCP server.

2020 – 2023

Senior Software Engineer

Jarvis Analytics · Dallas, TX (remote)

SaaS analytics platform for dental practices and multi-location dental organizations. Backend services, data workflows, integrations, and user-facing reporting across multiple practice-management systems; production reliability, data accuracy, and performance at scale.

2015 – 2020

Senior Web Application Developer

Salary Board Pte. Ltd. · Singapore (remote)

Labor-market analytics and compensation-data platform. Salary research tools, dashboards, and reporting used by professionals, companies, and government organizations; external data integrations and internal APIs through multiple growth stages.

2011 – 2015

Senior Software Developer

REfindly · Naples, FL (remote)

Web software for real estate agents, teams, and brokerages: responsive sites, CRM features, lead generation, and property search on listing-data integrations — across many branded installations with distinct configurations.

2006 – 2011

Freelance Web Developer

Self-employed · clients in Ireland & California

Business websites, database-driven PHP/MySQL applications, CMSs, and internal tools — owning the full lifecycle from requirements to ongoing support.

The composite picture

drawn from the systems above
01

Backend

  • — Laravel 11/12 · PHP 8.2–8.4 · primary stack
  • — Eloquent at scale · 361 models in one system
  • — Event-driven domain design · service + action layers
  • — Queued processing · Horizon · Redis
  • — Sanctum APIs · Reverb websockets · auth + realtime
  • — Node.js · NestJS · typed, modular web-app backends
02

Admin platforms

  • — Filament v4 & v5, in depth · resources, pages, widgets, relation managers
  • — Multi-panel setups · separate panels, one domain model
  • — Exporters & custom analytics · 18+ analytics pages built
  • — Major-version upgrades on live systems · Filament v4→v5 · Livewire 3→4 · Laravel 11→12
03

Frontend

  • — Next.js 16 App Router · React 19 · Zustand · SWR
  • — Bespoke design systems on Tailwind · Atomic Design, token layers
  • — Storybook + a11y checks · Playwright · Vitest browser mode
  • — Vue.js · Angular · production experience
  • — Livewire · Alpine.js · the TALL stack
  • — Legacy stacks · CodeIgniter · jQuery — maintained and modernized
04

Domains

  • — Double-entry accounting · AR, AP, journals, reconciliation
  • — Philippine statutory payroll · SSS · PhilHealth · Pag-IBIG · labor law
  • — Multi-branch inventory & procurement · reconciling, audited
  • — Subscription billing · commissions · renewals, edge cases
  • — Construction costing · BOQ / POW / cost-to-finish
05

AI engineering

  • — LLM content pipelines · auditable, reproducible prompts
  • — Classification over real corpora · churn & topic analysis
  • — Tool-calling agents · authorization enforced by the tool surface
  • — Agent orchestration · Claude Code: planner, executor, verifier and reviewer lanes
  • — Custom skills & MCP servers · team standards the agents can query
  • — Also fluent in · OpenAI Codex · Cursor · agent-context docs per subsystem
06

Infrastructure

  • — Laravel Forge · Laravel Cloud · provisioning & deploys
  • — DigitalOcean · AWS S3 · droplets, Spaces, object storage
  • — Ubuntu · Nginx · Supervisor · Redis · production operations
  • — Git · GitLab & GitHub CI/CD · automated pipelines
07

Integrations

  • — OpenAI · Meta Graph API · publishing, sync, webhooks
  • — Loyverse POS · Brevo · AWS S3 · typed API clients
  • — Headless-Chrome rendering · PDF generation · Browsershot · DomPDF · FPDF
08

Quality

  • — PHPUnit · Pest · 240+ test files at peak
  • — Playwright end-to-end · agent-driven browser tests on every ticket
  • — Larastan / PHPStan · baselines as a ratchet
  • — Pint · Rector · Telescope · formatting, upgrades, observability
  • — Audit trails & policies · on every financial record
09

Practice

  • — Technical leadership · review + merge authority, team of five
  • — Written architecture docs · a first-class deliverable
  • — Honest debt management · named, contained, scheduled