Marcelo Pereira Barbosa

Intro

I spent 26 years building and maintaining telecom billing systems — invoicing, tax, accounting and financial reconciliation, and subscriber base migrations. It is a domain where the software answers for money already charged and obligations already filed: correctness is not a quality detail, it is the product.

Over the past year I brought that discipline to development with AI coding agents. I use Claude Code, GitHub Copilot and Cursor to write most of the code; my job is to decide the architecture, define what has to hold true, and verify every change before it reaches production. Both live products were built this way — with unit tests, end-to-end Playwright specs, and deploy pipelines with a rehearsed rollback.

The underlying work has not changed: it is still engineering supervision. What changed is how fast the code appears, not who is accountable for it.

Experience

Projects

Software built by directing AI coding agents — Claude Code, GitHub Copilot and Cursor. The architecture, the acceptance criteria and the verification before production are mine.

In production

• Portal Brasileirão — Companion app for the Brazilian Série A championship

  Standings, fixtures and statistics for the Série A. OAuth login, Material Design 3 tokens and a live /api/health endpoint. Seven GitHub Actions workflows including rollback, with a documented flip-back drill, and keyless deploys over OIDC. When no API token is available the app falls back to a frozen snapshot instead of failing. React 19 · TypeScript · Express 5 · Tailwind 4 · AWS · 45 unit tests and 30 end-to-end Playwright specs · 742 commits in nine days (23–31 August 2026)

• Agora na Copa 2026 — Companion app for the FIFA World Cup 2026

  Monte Carlo simulator for qualification odds, using a Dixon-Coles model and the Article 13 FIFA tie-breaking criteria applied in the correct order. External integrations (FIFA API, Open-Meteo, Google Trends) each have their own fallback. Served by nginx and systemd. React 19 · TypeScript · Express · AWS · 34 unit tests and 37 end-to-end Playwright specs · 1,279 commits

Experiments and tooling

• ai_workflow — Verification pipeline for AI-assisted development

  Automates the step that usually gets skipped: making what an AI coding agent produced verifiable before it ships — review, documentation and tests.

• gitx — Git TUI wired up over MCP

  Terminal interface for Git commands, integrated with assistants through the Model Context Protocol (MCP).

• olinda_copilot_sdk.ts — TypeScript wrapper for the GitHub Copilot SDK

  A TypeScript layer over the GitHub Copilot SDK for programmatic use in custom tooling.

• Onde Estou? — Real-time geolocation

  Shows your current position on a map using the browser's Geolocation API, no registration required.

• MapaSP — Interactive map of São Paulo

  Interactive map of São Paulo city with points of interest and neighborhood navigation.

• Music in Numbers — Spotify data analysis

  Visualizes and analyzes your Spotify streaming history with charts and statistics.

• Monitora Vagas — Hotel vacancy monitoring

  Monitors hotel room availability and notifies you when openings appear.

• Libraries published on npm — bessa_patterns.ts, ibira.js and paraty_geocore.js

  Three packages published to the npm registry, still at early versions.

• Automation scripts — Utility scripts on GitHub

  A collection of utility scripts for everyday development tasks, available on GitHub.

The public GitHub portfolio holds more than 50 non-fork repositories.

About

People who understand tax rules, invoicing and reconciliation usually do not build modern software. People who build modern software have rarely reconciled a credit-card system. I spent 26 years on the first side and the past year shipping production products on the second.

In practice that means being able to say why a number is wrong before opening the code — and then fixing the code. It is the combination that matters in billing, reconciliation, revenue assurance, data migration and systems in regulated industries, where the cost of being wrong is not an ugly screen: it is an incorrect charge or an incorrect filing.

Outside of code, I am interested in music — I have a project to compose using Strudel.

Contact