Gabriel – Next.js, Typescript, React

A Claude Code plugin that installs a personal AI assistant running 24/7 for non-technical professionals such as lawyers and doctors. Each assistant has a name, a personality, long-term memory and its own routines, and it focuses on second-brain knowledge management and document drafting. The owner talks to it through Telegram or a web chat panel using text, audio, images or files. It runs on Claude Code Channels as a persistent session kept alive by systemd, so replies start in about one second with the system prompt cached, and it uses the owner's Claude Pro/Max plan with no per-message API cost. Launched as a paid product with real sales and a paid mentorship program where students implement it. Currently at version 0.9.6.

• Designed the runtime: one live Claude Code session per agent, managed by systemd with about 13 timers and strict sandboxing, plus a watchdog that restarts stuck sessions or expired logins and a nightly audit;
• Built a layered memory system: daily working memory, long-term memory, journals, relationships and goals, keyword recall through a UserPromptSubmit hook, optional semantic search with Gemini or OpenAI embeddings stored in SQLite, and a nightly reflection where the agent consolidates its own memory;
• Built a multi-agent team feature: specialists created by the owner become native Claude Code subagents, the main agent orchestrates them, and PreToolUse hooks limit each one to its own tool catalog (9 of 10 correct delegations in evaluation);
• Wrote a custom MCP server for the web chat panel (JSON-RPC 2.0 over stdio, zero dependencies) and integrated the official Telegram MCP channel, with audio transcription through Groq, OpenAI, ElevenLabs or a local model;
• Wrote the core in Bash and standard-library Python with no pip installs, including a .docx generator, an .ics parser, a knowledge graph and AWS SigV4 signing for S3/R2 backups;
• Built the Node.js/Express admin panel in plain JavaScript, with password auth, rate limiting and localhost-only binding by default;
• Designed tool access so the agent never holds credentials: scripts keep the tokens and refuse anything not allowed (Trello, BrasilAPI, calendar, documents, backups), plus about 20 skills for setup, onboarding interview, second brain, documents, calendar, diagnostics and updates;
• Set up quality and releases: about 64 test suites in Bash, Python and Node with mutation testing, 30 QA and security reports, a Docker lab running Debian with real systemd (also tested on aarch64 with 2 GB RAM), versioned ZIP releases and an update command with rollback;
• Shipped three deployment modes (Lite, Local and 24/7 Server) and run two production instances on a Raspberry Pi.

A web app that gives Claude users a library of ready-to-use prompts and materials adapted to their own business. A short onboarding of 11 questions (4 required, 7 optional) captures the offer, the ideal customer and the goals, and those answers become the context injected into every prompt, so Claude writes about the user's real business instead of generic text. Each answer is saved as soon as it is given, so users can stop and come back later.

• Built the product end to end in Next.js, deployed on Vercel;
• Designed the onboarding flow and the context model that turns the user's answers into prompt context;
• Wrote and organized the prompt library by business area, with templates that receive the user's context automatically;
• Built it with an agentic workflow in Claude Code, using subagents for code review and QA.

A members-only luxury subscription club (golf, bespoke tailoring, wine, watch and jewelry curation, concierge and private experiences) wanted to run sales and member service with AI. Gabriel designed an ecosystem of 11 AI agents plus a CRM and dashboard where the CEO controls them, and delivered a working MVP with three active agents: customer service, sales and business intelligence. WhatsApp messages are classified, routed to the right agent, answered, logged in the CRM and moved through the sales pipeline in real time. Email replies are drafted by the agents and wait for human approval before sending.

• Designed the architecture for an 11-agent ecosystem and delivered the MVP with the customer service, sales and intelligence agents active;
• Built the end-to-end WhatsApp flow on Z-API: classify, route, reply, update the CRM and the pipeline, with live dashboard updates over Server-Sent Events;
• Built the Gmail flow where agents draft replies for human approval and send daily digests;
• Gave the CEO direct control over the agents: editable guidelines per agent injected into the prompts, AI conversation summaries and AI lead analysis;
• Built the CRM and dashboard in React with Recharts, including subscription KPIs such as MRR, ARR and members per plan;
• Implemented authentication with three roles (attendant, CEO and super admin) using scrypt and HMAC from node:crypto, with no external auth libraries;
• Built an idempotent WhatsApp contact import with phone-number deduplication, following LGPD rules;
• Added built-in simulators for every integration, so the product runs end to end before the client connects real WhatsApp, Gmail and OpenAI credentials;
• Wrote the technical README and a UX brief so Claude Code agents can keep developing the product from the documentation.

A mobile expense tracker for iOS and Android, built on the idea that writing down each expense by hand helps people spend less. Logging an expense takes under five seconds, works offline and syncs when the connection returns. It supports BRL, EUR and USD side by side, categories and cards, a home screen widget, and a shared account where couples see each other's expenses in real time after joining with an 8-character invite code. The business model is free with discreet ads, plus a one-time premium purchase instead of a subscription.

• Built the React Native app for iOS and Android and the NestJS API, self-hosted on a Raspberry Pi;
• Designed the offline-first data layer, with local storage and sync when the device reconnects;
• Built shared accounts with real-time updates, invite codes and personal cards kept private inside a group;
• Implemented multi-currency support, a home screen widget for quick entry, ads and the one-time in-app premium purchase;
• Developed the app with Claude Code agents from written specs, with subagents for tests and code review;
• Published the app on the App Store and Google Play.

A certificate issuing platform built for the Brazilian market. Global certificate tools are priced well above what Brazilian customers pay, and local event platforms treat certificates as a side feature, so Certifizi focuses on certificates as the core product, for solo creators up to large companies. The business model combines subscription tiers (Free, Starter, Pro and Business) with pay-per-event pricing based on the number of participants. Gabriel built the product solo with React, TypeScript and NestJS, using Claude Code agents throughout development.

• Built the React and TypeScript front-end and the NestJS API;
• Ran the market research and defined the pricing strategy, with a freemium entry, subscription tiers and pay-per-event plans;
• Built AI-assisted features for certificate creation, using LLMs to generate certificate text from the event data;
• Developed the product with Claude Code agents from written specs, with subagents for tests and code review;
• Wrote the product and strategy documentation in English and Portuguese.

A multilingual crossword game based on the Brazilian "diretas" format, where arrows in the dark cells point to where each answer goes, in four directions. It works as a learning tool for English, Spanish, German, French and Italian, teaching vocabulary and correct word boundaries. Players earn XP and levels, keep streaks, unlock achievements and compete in daily challenges and leaderboards. Gabriel built it solo as a full-stack product, using an agentic development workflow in Claude Code and LLMs to generate and review the vocabulary and clue content.

• Designed a three-phase puzzle generation algorithm that fills a 16x11 grid with about 45 words;
• Built the React and Tailwind front-end and the NestJS API with Prisma, PostgreSQL and Firebase authentication with Google login;
• Built the gamification layer: XP and levels, streaks, achievements, daily challenges, leaderboards and player analytics;
• Built an LLM pipeline that generates and reviews vocabulary and clues across five languages;
• Developed with Claude Code agents from written specs, with subagents for tests and code review;
• Implemented role-based access control (user, admin and super admin), Swagger API docs and health checks;
• Containerized front-end and back-end with Docker multi-stage builds and nginx.

A reading tracker for the Brazilian market that Gabriel built solo, from architecture to production. Readers log their progress, organize books into lists and see statistics about their reading habits. It ships as a React web app and a native Android app on a shared NestJS REST API, self-hosted on a Raspberry Pi 5 with a full CI/CD pipeline. The project is built with an agentic workflow: Claude Code agents implement features from written specs, and custom skills, hooks and subagents handle code review, tests and releases.

• Designed the front-end and back-end architecture and the Prisma/PostgreSQL schemas;
• Built the CI/CD pipeline that auto-deploys to a Raspberry Pi 5, with Grafana and Prometheus monitoring;
• Set up an agent-driven development workflow in Claude Code, with spec-first tasks, custom skills, hooks and subagents for code review, test generation and release notes;
• Built AI-assisted features on top of the reading data, such as book metadata enrichment and reading insights generated by LLMs;
• Wrote a LinkedIn article about the technical decisions and real bugs on the way to the Play Store.

Long-running engagement with a Brazilian sports network, covering front-end products, marketing automation and web operations. The main product was a custom calendar for the Paris 2024 Olympics listing every sport, with filters by sport, date and medal events and export to Google Calendar and iCal. Gabriel also runs the HubSpot automations and the WordPress/WooCommerce sites. Since 2025 the work includes AI: LLM-based automations for content and CRM data, and development done with Claude Code agents.

• Architected the front-end from scratch: a responsive, high-performance React interface for the Olympic calendar;
• Built search and filtering by sport, date, and medal events for smooth navigation across the full event schedule;
• Integrated Google Calendar and built an iCal export so users could save events on iOS, Mac, and any calendar app;
• Delivered HubSpot marketing automations and campaigns, and maintained multiple WordPress sites.
• Built LLM-powered automations that draft campaign content and enrich lead data inside HubSpot workflows;
• Moved day-to-day development to an agentic workflow with Claude Code, using agents for WordPress maintenance, plugin updates and QA.

Product and front-end lead for an autism education platform that has served nearly 40,000 students since 2018. For 8 years Gabriel has owned the front-end architecture and technical direction as it grew into a full learning ecosystem: live classes, community forums, downloadable materials and satellite apps unified by single sign-on. He also leads the platform's AI work, including a Portuguese support agent with RAG over the course catalog and Meus Progressos, an ABA assessment app for clinics, therapists and parents built with an agentic Claude Code workflow.

• Owned the front-end architecture and led the platform's technical direction for 8 years as it scaled to nearly 40,000 students;
• Led a distributed team of in-house and freelance engineers, driving both product management and hands-on development;
• Designed a scalable Styled-components design system powering a large, evolving React + TypeScript codebase, documented with a Storybook component library;
• Architected an ecosystem of satellite services and apps unified through single sign-on (SSO);
• Built and maintain the AWS deployment pipeline (S3 + CloudFront) with automated CI/CD via GitHub Actions;
• Designed and built a Portuguese AI support agent for students: a router/classifier sends each question to one of three specialized prompts (platform navigation, replies to course comments, and RAG over 268 articles, 125+ courses and 209 materials), tuned for an audience mostly over 45 and for misspelled technical terms;
• Built Meus Progressos, an ABA (Applied Behavior Analysis) assessment app with role-based access for clinics, therapists and parents, using NestJS, Prisma, PostgreSQL and React Native;
• Adopted spec-driven development with Claude Code agents, using Figma MCP for design tokens and Playwright MCP for automated platform mapping;
• Integrated OpenAI-powered features into the platform.

A digital marketing agency focused on online course launches, paid ads and sales. Gabriel has handled everything IT-related since the agency started: servers, domains, email, landing pages, and the automations and integrations between platforms. He runs a self-hosted n8n instance on Docker that connects the main sales platforms (Hotmart and Stripe) through APIs and webhooks, updating the CRM, Supabase and ActiveCampaign, recovering abandoned sales and sending WhatsApp notifications. Since 2025 the automation layer includes AI agents in n8n that qualify leads, answer students on WhatsApp through Evolution API and summarize sales data for the team, and new internal tools are built with Claude Code.

• Developed pages using Elementor + Wordpress utilizing best practices for maximum performance;
• Adapted plugins for Wordpress;
• Managed Servers on Digital Ocean/AWS/Shared Hosting Solutions;
• Managed DNS;
• Developed landing pages with HTML/SASS/VueJS;
• Created and maintained automations with N8N, Make and Manychat to improve the team and company efficiency, sometimes by a factor of 10;
• Gathered data from traffic and sales into unified databases to facilitate queries and to guarantee no data gets lost, ever;
• Implemented conversion and tracking APIs;
• Built AI agents in n8n with OpenAI and Claude for lead qualification, WhatsApp support through a self-hosted Evolution API and daily sales summaries;
• Ran large WhatsApp outreach campaigns with rate control and lead state tracked in n8n data tables;
• Managed a tech team.

An automation platform for a staffing company in skilled trades and construction. It turns incoming job openings into recruitment campaigns across SMS, Facebook groups and LinkedIn without manual work. The system polls ADP Workforce Now for new job requisitions, removes duplicates and extracts trade, location, pay and description from each one. LLM agents (OpenAI) parse client intake emails from HubSpot into structured job orders and write the campaign copy from real job data. A geo and skill matching API finds qualified workers near each job site. Operators use a web dashboard to monitor runs, trigger campaigns and edit matching rules, while a cron engine runs the pipeline during business hours. Built on Node.js/Express, PostgreSQL, n8n and Playwright, fully Dockerized with CI/CD.

• Designed and built the backend architecture from scratch using Node.js/Express and PostgreSQL;
• Developed an ADP Workforce Now integration with OAuth 2.0 and client-certificate auth, including an incremental job-requisition polling pipeline with deduplication and recoverable polling state;
• Built a multi-channel campaign distribution engine that automatically posts to SMS, Facebook groups and LinkedIn from a single job record;
• Integrated a geo/skill worker-matching API using AWS SigV4 request signing to surface qualified candidates near each job site;
• Built an LLM agent (OpenAI GPT-4o-mini, orchestrated in n8n) that reads client intake emails and extracts company, skills, pay rates, timelines and job site details into structured records;
• Implemented LLM-generated campaign copy, so every SMS and social post is written from real job data with prompt templates per channel;
• Created a database-driven trade-matching system with flexible keyword extraction, exclusion rules and full CRUD management via the dashboard;
• Built the operations dashboard for real-time monitoring, manual campaign triggers and automation configuration;
• Developed a cron-based scheduler that runs the full pipeline within configurable business operating hours (timezone-aware);
• Hardened public webhook endpoints against SQL injection, added rate limiting, URL/input validation and optional signature verification;
• Containerized the whole stack with Docker and set up a GitHub Actions CI/CD pipeline with deploy webhook notifications;
• Wrote and maintained a comprehensive automated test suite (500+ Jest unit/integration tests plus Playwright E2E coverage).
• Used AI-assisted coding with Claude Code throughout the project, from scaffolding and refactors to writing and maintaining the test suite.

The project was a global marketplace designed to enable Brazilian sellers to reach international customers. The platform focuses on optimizing conversion rates across various currencies, allowing users to sell both digital and physical products. The goal is to provide a seamless experience for sellers looking to expand their reach while ensuring competitive and efficient transactions for buyers worldwide.

• Designed and implemented back-end architecture for internal-facing API;
• Developed and managed front-end architecture using React/Next.js;
• Configured multiple deployment environments on AWS Beanstalk;
• Set up Lambdas to offload tasks from servers;
• Configured and managed multiple SQS queues for handling incoming orders;
• Led the development team, ensuring alignment with company goals while guiding their individual focus areas.

Developed a platform for a startup aimed at identifying top talent for various open job opportunities. The project focused on streamlining the recruitment process, enhancing candidate sourcing, and matching the best-fit candidates to client needs, all while providing an intuitive and efficient experience for both recruiters and candidates.

• Created a design for the customer intake form, focusing on user experience and ease of use;
• Developed the architecture and front-end for the customer intake form, ensuring scalability and responsiveness;
• Designed a new HRMS tailored to the company’s specific needs, enhancing workflow efficiency;
• Implemented automated tests for Vue and React/TypeScript projects to ensure code quality and reliability.