About

Background

Staff Software Engineer at Jusbrasil with over a decade in legal tech, across data science, backend, tech lead, and engineering management roles. Since 2025 I have been hands-on again, building and operating LLM agents in production. PhD in Computer Science (social network analysis), with publications on expertise and trust in online communities.

Two of the products I led went from zero to revenue and are still growing. The background-check product I launched has tens of thousands of subscribers today, several times what it had when I handed it off. The WhatsApp legal-intake channel serves over 100,000 people a month and has produced hundreds of thousands of leads, and leads enriched by the agent sell at a materially higher price than unenriched ones. The agent platform, the evaluation service, and the observability work described below were built to keep those products growing.

From expert finding to LLM-as-judge

My thesis and most of my publications ask the same question in different forms: in a community of people asking and answering questions, who answers well, whom can you trust, and which discussions are worth your attention. I spent ten years building metrics and models to answer that for humans.

Today I build automatic judges that evaluate whether an agent answered a legal question well. It is the same research problem with the answerer replaced by an LLM: define what a good answer looks like, measure it at scale, and use the measurement to decide what ships. The evaluation service, the synthetic users, and the daily evals as a deploy gate are that thesis applied to production.

What I do well

  • LLM agents in production: multi-agent prompt design, A/B experiments across prompt and model versions, channel-agnostic chat APIs, migration between agent platforms.
  • Evaluation: LLM-as-judge pipelines, synthetic users, regression evals, quality metrics as first-class telemetry.
  • Observability and cost: Prometheus and Grafana, per-turn token and latency accounting, alerting on product signals.
  • Platform: Python (FastAPI), TypeScript (React, Next.js monorepo), Go, Postgres, Kubernetes, Terraform, GCP and Vertex AI.
  • Leadership: technical roadmaps for generative AI, mentoring engineers into tech-lead roles, hiring for senior positions, written communication as the default.

Products I took to market

  • Technical lead for the background-check product, from launch to the numbers above, covering product management alongside engineering.
  • Technical lead for the “Seus Direitos” acquisition channel on WhatsApp, from the first bot to the chat platform described in Now, including the commercial negotiation with Meta.

Earlier data work

  • Legal entity detection engine behind normative-acts citations, foundational for Jusbrasil search.
  • Cleanup and deduplication of the laws corpus and reprocessing of citation links at scale.

Selected publications

  1. How Do Outstanding Users Differ From Other Users in Q&A Communities? (2019)
  2. Trust Investigation in Communities Using Feature Learning (2019)
  3. Modelling Experts Behaviour in Q&A Communities to Predict Worthy Discussions (2017)
  4. Finding Topical Experts in Question & Answer Communities (2016)
  5. How to find people who can help to answer a question? Analyses of metrics and machine learning in online communities. Computers in Human Behavior (2014)

All publications

International

  1. How Do Outstanding Users Differ From Other Users in Q&A Communities? (2019)
  2. Trust Investigation in Communities Using Feature Learning (2019)
  3. Learning in Communities: How Do Outstanding Users Differ From Other Users? (2018)
  4. Experts and likely to be closed discussions in question and answer communities: An analytical overview (2018)
  5. Modelling Experts Behaviour in Q&A Communities to Predict Worthy Discussions (2017)
  6. Finding Topical Experts in Question & Answer Communities (2016)
  7. How to find people who can help to answer a question? Analyses of metrics and machine learning in online communities (2014)
  8. Finding Experts on Facebook Communities (2014)
  9. Finding Reliable People in Online Communities of Questions and Answers: Analysis of Metrics and Scope Reduction (2014)
  10. Finding Reliable Users on Online Communities Using Artificial Neural Networks (2014)

In Portuguese

  1. Aprender ao longo da vida através de ambientes pessoais de aprendizagem (2021)
  2. Captura da Aceitação do Blackboard e do Tipo de Motivação de Alunos de Cursos Presenciais de Ciências Exatas em uma Universidade Privada (2020)
  3. Coautor do capítulo “Da Pesquisa Científica à Inovação” do livro “Pesquisa e Inovação: Visões e Interseções” (2017)
  4. Prospecção Tecnológica: Levantamento de Patentes, Atuação da Academia e Potenciais Inovações em Ambientes de Aprendizagem no Brasil de 2000 a 2015 (2016)
  5. Exploratório das Produções e Colaborações entre Pesquisadores em Informática na Educação: uma Análise de Publicações do SBIE de 2001 a 2013 (2015)
  6. Análise Empírica de Comunidades Online baseadas em Enriquecimento Semântico para encontrar Usuários Especialistas (2015)
  7. Identificando Colegas para Ajudar em Minhas Dúvidas: Um Estudo Empírico em Comunidades de Perguntas e Respostas (2015)
  8. Encontrando Usuários Confiáveis em Comunidades Online de Perguntas e Respostas Através de seu Índice de Confiança (2014)
  9. Importância dos Testes Automatizados e da Integração Contínua no Uso do Scrum (2012)
  10. ZK Framework: Utilizando Ajax sem Javascript. Revista Mundo Java, edição 36 (2009).
  11. Projeto Integra: Uma Proposta de Desenvolvimento de um Ambiente de Colaboração Integrado ao SIGA e aos Serviços Oferecidos pelo Google (2008)