Hugo Barbosa
ENPTES中文

Hugo Barbosa · Tech Lead

I build AI products that hold up in production.

Tech Lead at Greenn, a Brazilian payments and online courses platform, working remotely from Brazil (UTC-3). LLM pipelines, resilient queues and agentic engineering.

Get in touchSee the work

What I bring

  • Technical leadership.From architecture to delivery, with teams that decide well even when I am not in the room.
  • AI in production.LLMs, RAG, voice and video pipelines serving real users.
  • Cost under control.Architecture and infrastructure reviews that mapped up to 50% in savings.
  • Since 2018.Shipping software to production, from freelance systems to SaaS.

Selected work

Bluee

greenn.com.br/bluee

An AI platform that turned a topic into a complete online course, with lessons, narration and an avatar presenter.

Role: Back-end engineer, then Tech Lead. 2025 to 2026.

1 hour from a topic to a complete AI-generated course

  • LLM layer with streaming and provider fallbacks, plus an agentic chat with parallel tool calls.
  • RAG over the user's own documents.
  • Avatar video pipeline: lip-sync, image-to-video and cloned voice, orchestrated through queues.
  • Resilient jobs with backoff retries, idempotency checks and delayed queues.
  • Laravel 12
  • PHP 8.4
  • Python
  • React
  • RabbitMQ
  • Redis
  • PostgreSQL

Greenn Club

greenn.com.br/club

The members area where course creators host and deliver their courses to thousands of students.

Role: Tech Lead. 2026.

R$20k+ saved per month on video storage

  • Verifiable certificates signed with ECDSA, a public key set and QR validation anyone can check.
  • Fair dispatch: per-tenant scheduling so one large creator never starves the queue for everyone else.
  • Subtitle pipeline moved from polling to webhooks, hardened against concurrent deliveries.
  • Legacy webhooks moved into the domain layer behind characterization tests.
  • Laravel 11
  • Vue 3
  • Redis
  • RabbitMQ
  • MySQL
  • S3
  • PandaVideo

PR Review Panel

Internal tool, no public link

My own tool for reviewing pull requests in depth: six phases run by Claude Code agents, in parallel, on every PR.

Role: Creator. 2026.

260+ pull requests reviewed with the panel in 2026

  • Six phases (quality, code, coupling, over-engineering, tests and team standards), each returning findings in a strict JSON schema.
  • An adversarial agent tries to refute every blocking finding, and Jev, a decision model on OpenRouter, triages the rest.
  • Agents navigate the code through a Graphify knowledge graph, with closed tool lists, read-only access and isolated git worktrees.
  • Every design choice is measured.
  • Next.js
  • TypeScript
  • SQLite FTS5
  • Claude Code
  • OpenRouter
  • Graphify
  • GitHub CLI

Experience

  1. 2026 to now

    Tech Lead · Greenn

    • Teams of up to 6 developers across Bluee and Greenn Club.
    • I set the architecture and code standards.
    • Cost reviews that mapped up to 50% infrastructure savings.
    • Security issues caught in review before they reached production.

  2. 2025

    Back-end Developer · Greenn

    • Back-end engineer on the AI features of Bluee and the reliability of its job queues.

  3. 2024 to 2025

    Full-stack Developer · Betabit

    • Led the AI chat of a real-estate CRM used by 15 agencies and hundreds of brokers: as early as 2024, agents queried the system and took actions by text and voice, on web, mobile and WhatsApp. Gemini, RAG with PGVector and a Redis cache answering in under 100 ms, for about 3,000 requests a day.
    • Led the AI insights module of a real-time marketing dashboard, built with Laravel Reverb and a Python ETL service.
    • TDD with Pest, code review on every PR and weekly deploys. Docker and GitHub Actions cut developer onboarding from hours to minutes.

  4. 2022 to 2024

    Web Developer Intern · UNIMONTES

    • Finished and launched the university's extension management system (Laravel, MySQL, SQL Server), adopted by every extension department.
    • Refactored the queries that blocked the system at enrollment peaks, and built a micro-certification module presented at CIEI 2024.
    • Introduced Git Flow and PR standards, mentored an intern and taught a Laravel workshop.

  5. 2021 to 2022

    Research Scholar · UNIMONTES

    • Modeled the new graduate admissions system with the graduate programs: requirements, data model and architecture.
    • Left a continuity guide used by the next scholars, and a paper at FEPEG 2022.

  6. 2017 to 2021

    Early years

    • IT technician at IFNMG, then freelance web development and computer repair as a sole proprietor.
    • First systems in production: a website for a solar energy company and an offline point of sale for a local shop.

How I work

Skills

  • LLMs
  • RAG
  • Claude Code
  • Embeddings
  • pgvector
  • OpenRouter
  • Gemini
  • MCP
  • Codex
  • Cursor
  • Jev
  • Graphify
  • BERT
  • N8N
  • Laravel
  • PHP
  • Python
  • Java
  • Spring Boot
  • RabbitMQ
  • Redis
  • PostgreSQL
  • MySQL
  • SQLite
  • Node.js
  • Laravel Octane
  • Laravel Reverb
  • SSE
  • Webhooks
  • TypeScript
  • React
  • Vue
  • Next.js
  • Inertia
  • Quasar
  • Astro
  • Expo
  • Three.js
  • GSAP
  • Docker
  • AWS
  • GitHub Actions
  • Datadog
  • S3
  • Cloudflare
  • Supabase
  • Linux
  • TDD
  • ADR
  • Pest
  • SDD
  • DDD
  • Clean Architecture
  • SOLID
  • KISS
  • YAGNI
  • CI/CD
  • Git Flow
  • ECDSA
01Guardrails before autonomy.

AI agents get closed tool lists, strict output schemas and a reviewer that argues back.

02Built for production, priced for scale.

Architecture that holds real load without the cloud bill growing faster than the user base.

03Trade-offs in the open.

Every technical decision comes with the options I rejected, what each one costs and what could go wrong, written down as an ADR.

04Uncertainty first.

Spikes, prototypes and real numbers come before commitments, so estimates stop being guesses.

05Security and quality from the first commit.

Threat checks, tests and code review are part of done.

06The simplest solution that works.

Occam's razor for software: fewer moving parts break less and cost less at scale.

07A team that keeps learning.

I review to teach, write decisions down and pair with people, so the team decides well when I am not in the room.

Toolbox

Claude Code, Codex and Cursor every day, with prompt engineering, harnesses and guardrails. Jev (a decision model), Graphify (a knowledge graph of the code) and Ponytail (a guard against over-engineering) for defensive automated review.

Learning

TechLead de Impacto mentorship with Alexandro Hervis. Postgraduate in Software Engineering for Applied AI at UniPDS, with professors such as Erick Wendel. Reading: Clean Code, The Clean Coder, The Pragmatic Programmer, Tidy First?, Software Architecture: The Hard Parts.

Independent projects

Own SaaS

Scanlex and Timesheet Transcriber AI (2025)

A SaaS that digitized timesheets and payslips for payroll specialists, with paying subscribers, and its free open-source version: bring your own Gemini key, crop pages, export to CSV or XLSX. Laravel, Vue, React, Gemini.

timesheet-transcriber-ai.vercel.appGitHub

For a client

KM Topografia

HR document and fleet management for a land surveying company, in production. Expiry alerts by email and WhatsApp, S3 with a local cache, audited roles. I designed the architecture and the infrastructure. Laravel 12, PostgreSQL, Docker. Private repository.

s.kmtopografia.com.br

For a client

Mia (2025)

WhatsApp receptionist for a health clinic: leads, scheduling and reminders. N8N, Gemini, Evolution API.

Personal

anthrya

Health app built AI-first: Bluetooth body scale, diet, glucose, workouts and body measurements, all reachable by AI through an MCP server. Raw readings stay separate from estimates, so history can be recalculated. Expo, Supabase, Hono. In development, private.

Education

  1. Lato sensu postgraduate program in Software Engineering for Applied AI.UniPDS, 2026 to 2027.
  2. B.Sc. Information Systems.UNIMONTES, 2021 to 2025. Thesis on BERT models for bug localization.
  3. IT Technician.IFNMG, 2017 to 2019.

Research projects

  1. Improving the accuracy of traceability recovery between software artifacts with generative AI.UNIMONTES, 2024 to 2026.
  2. Undergraduate research (PIBIC/FAPEMIG): a system for managing selection processes at UNIMONTES.2021 to 2022.
  3. Volunteer undergraduate research (ICV): residential installations with solar micro-generation.2020 to 2021.

Publications

Titles translated; papers in Portuguese.

  1. Exploratory analysis of BERT-based models for software bug localization.ENMC 2025
  2. Performance of pre-trained embedding models for bug localization on ArgoUML.CIEI 2025
  3. SBERT for software bug localization.CIEI 2024
All 7 publications
  1. Micro-certification module for a university extension management system.CIEI 2024
  2. Impact of the Internet of Things on quality of life at home.CBICT 2022, honorable mention
  3. A web application for graduate admissions.FEPEG 2022
  4. Performance of a photovoltaic plant from automatically collected data.FEPEG 2020
  5. Comparative analysis of proprietary embedding models for bug localization.Manuscript (unpublished), 2025

Languages

  • Portuguesenative
  • Englishadvanced
  • Spanishintermediate
  • Frenchbeginner

Want to talk about AI, architecture or leadership?

Send emailLinkedInGitHub