Overview

Jorge Castro is a multilingual SaaS that acts as an AI-driven SEO department for a business website. A brand connects its site, and the platform audits it, finds keyword opportunities, writes and publishes content, then tracks Google rankings and how often the brand shows up in AI answers. The web app is also packaged as an Electron desktop app, and a separate telesales app supports the sales team.

My role

I have been a core engineer on the platform since mid-2025 and am the most active committer on the repository. I own the AI Overview and AI visibility tracking features and the Electron desktop shell, and I work across the CastroLab tool suite, the rank tracker, prompt templates, content generation and change logging, the publishing integrations (I built the custom-website adapter), the dashboard UI and its translations, and the deploy pipeline.

What it does

• Site audit crawler that flags technical and on-page SEO issues
• Keyword research, rank tracking and Google AI Overview tracking
• AI visibility reports showing whether ChatGPT, Gemini, Perplexity and Claude mention or cite the brand
• Content generation for meta data, product and category pages, articles and press releases, with a log of every edit
• An autonomous agent that picks near-page-one keywords and runs content modules for each brand on a schedule
• Publishing to WordPress, Shopify and any custom backend

How it's built

The dashboard is a Next.js app using Redux, next-intl (eight languages) and Socket.io for live job progress. The backend is a Node.js and Express API on MySQL. Long-running work goes onto a Redis-backed Bull queue that separately scaled worker containers process, scheduled work runs in a cron worker, and heavy crawls run as Cloud Run jobs. An internal LLM gateway routes each request to OpenAI, Azure OpenAI, Anthropic, Gemini or other providers and estimates token cost. Crawling uses Crawlee with Playwright. Every service ships as a Docker image to Google Cloud Run through GitHub Actions.

Highlights

• One Redis job queue with retries, exponential backoff and stalled-job detection keeps long AI and crawl jobs reliable across many worker replicas.
• A provider-agnostic LLM gateway lets each feature choose a model per request and tracks cost per job.
• A pluggable publishing layer pushes the same content to WordPress, Shopify or a documented custom REST contract.
• Before AI visibility results are saved, each cited page is crawled and a brand is counted only where it really appears on that page, so one sentence citing several sources cannot inflate the mention count.