Back

About

I'm a Software Engineering student at the University of Brasília (UnB) and an AI Resident at LabLivre × Instituto Eldorado.

I design, build and ship software, with AI in the architecture and in the workflow.

As co-founder and CTO of Macronutri, I took a no-code prototype to a production SaaS in three weeks. I made the architecture calls; AI coding agents handled the implementation.

My background in UX keeps real users at the center of what I build.

I care about systems that are reliable, secure, and useful to people.

Download CV femathrl@gmail.com

Experience

  1. Co-founder & CTO

    Oct 2026 – present

    Macronutri · Remote

    • Took Macronutri, a SaaS for Brazilian dietitians, from a no-code prototype on Base44 to a product live in production in about three weeks.
    • Owned every strategic and architectural decision: stack, data model, security and privacy model, infrastructure, cost architecture and technical roadmap. AI coding agents carried out implementation under my specs and review.
    • Led UX for the product, including interviews and usability tests with dietitians.
    • Worked with the co-founder/CEO on product scope, positioning and go-to-market.
    Case study

  2. AI Resident & Researcher

    Sep 2026 – present

    AI Residency, LabLivre (UnB) × Instituto Eldorado · Gama, DF

    • Selected for an 8-month, in-person residency in applied AI and data science, run as Challenge Based Learning: real-world challenges, 5-person teams, weekly sprints and public showcases.
    • Led a 5-person team on Dossiê, a multi-agent system that builds evidence dossiers for suspicious news. Coordinated people and sprints, defined the architecture, and orchestrated the agents into one pipeline.
    • Speaker at the "AI Beyond LLMs" workshop and co-lead of the "AI Workflows" workshop for the cohort.
    • Co-authoring a research paper with a residency mentor in the MLOps / LLMOps / AIOps space (in progress).
    Dossiê case study

  3. Data Science & Analytics Intern

    2026

    CNI (National Confederation of Industry), Strategy & Performance · Brasília, DF

    • Monitored national industry indicators with Python, Databricks and Power BI.
    • Built about three AI agents for the team. One cut the production cycle of technical reports for the Management Efficiency Program (PEG) from 2–4 weeks to 1–2 weeks.
    • Built and tuned a Databricks Genie (AI/BI) agent over Unity Catalog, running in production: instruction tuning, example SQL queries, trusted assets, human-in-the-loop review and periodic answer revalidation.
    • In the AI & AI Agents unit of the Data Office, led the creation of guidelines and a reference manual for implementing AI tools: design system, architecture, requirements elicitation and governance.
    • Masked Spark SQL queries over the corporate data lake to comply with LGPD, Brazil's data protection law.
    • Wrote Python scripts for value reconciliation and data checks, using AI-assisted development.

  4. UX Designer & Research Assistant

    ITRAC, Software Engineering Lab, UnB · Brasília, DF

    • Worked on UX for projects in collaboration with Brazil's Ministry of Management and Innovation in Public Services (MGI).
    • Led a 3-person team in a Systematic Literature Review on the state of the art of UX Design in Software Engineering. The team screened 1,500+ papers and produced a final report.
    • This foundation in research methods and user-centered design shaped how I designed Macronutri.

Decision log

Five calls from Macronutri. I made each decision; AI coding agents implemented it under my specs and review.

  1. Business rules live in the database, not the UI

    D-01
    Context
    In the no-code prototype, the free-plan patient limit and the subscription tier were enforced only in the browser, so they could be bypassed.
    Decision
    Move the free-plan limit into a Postgres trigger with locking, so concurrent requests can't bypass it. Every table has Row Level Security with per-dietitian isolation. SECURITY DEFINER functions run with a fixed search_path, privileges revoked and then granted explicitly.
    Check
    238 RLS integration tests run in CI against Supabase in Docker.

  2. A payment can't be cancelled by a race

    D-02
    Context
    Two simultaneous payment-gateway events could cancel the subscription of a customer who had paid.
    Decision
    The Asaas webhook is idempotent, validates signatures on the raw body, and uses optimistic concurrency.
    Check
    A real bug caught in review, then fixed.

  3. The AI assistant never sees patient names

    D-03
    Context
    Anna, the in-product assistant, answers clinical-practice questions over health data that is sensitive under LGPD art. 11.
    Decision
    Patients are pseudonymized (P1, P2…), and national ID, phone and email are masked. Clinical scope limits are enforced. A monthly AI quota is reserved in the database and refunded if the call fails.
    Check
    A live evaluation suite (eval:anna) tests the assistant's behavior.

  4. Near-zero fixed cost

    D-04
    Context
    An early-stage product needs hosting it can afford before revenue, under terms that allow commercial use.
    Decision
    Serve the frontend from Cloudflare Workers (Static Assets). Vercel Hobby was evaluated and rejected because its terms prohibit commercial use.
    Check
    Deployed to staging and production through the same reusable pipeline.

  5. Migration that respects deletion

    D-05
    Context
    Data had to move from the no-code prototype without bringing back patients deleted on an LGPD request.
    Decision
    A repeatable migration tool that runs as a dry run by default and never resurrects deleted patients. Deletion removes Storage files first, then the records.
    Check
    Review caught imports resurrecting deleted patients, and it was fixed.

Research & teaching

  • Research paper on operating and validating AI systems in productionIn progress
    Co-authoring with a mentor from the AI Residency, in the MLOps / LLMOps / AIOps space.
  • Speaker, "AI Beyond LLMs" workshop
    AI Residency. Reinforcement Learning vs. LLMs and when each fits, guardrails, and shielding for safe AI behavior.
  • Co-lead, "AI Workflows" workshop
    AI Residency. Skills, prompt engineering, subagent-driven development and practical AI-assisted engineering workflows.
  • Co-founder & instructor, Neural Networks study group
    Co-founded and ran a study group on neural networks for one year, teaching the fundamentals to participants.
  • Systematic Literature Review: UX Design in Software Engineering
    ITRAC, UnB. Led a 3-person team that screened 1,500+ papers and produced a final report.

Skills

Full-stack & cloud
TypeScript, React 19, Vite, Tailwind CSS, Supabase, PostgreSQL 17, Row Level Security, Edge Functions (Deno), pg_cron, Cloudflare Workers, Payment webhooks (Asaas)
DevOps & security
GitHub Actions CI/CD, Docker, Staging / production environments, Encrypted backups, Threat modeling, Penetration testing (OWASP ZAP), gitleaks, osv-scanner, Vitest, Playwright
AI engineering
Multi-agent system design & orchestration, Spec-driven development with AI coding agents (TDD, reviewer agents, subagents), Prompt engineering, LLM features with pseudonymization, guardrails and live evals, Databricks Genie tuning (instructions, example SQL, trusted assets, HITL)
AI development tools
Claude Code, OpenCode, Antigravity, Gemini
Machine learning
Neural network fundamentals, Reinforcement learning fundamentals, Guardrails & shielding
Data
Python, SQL (Spark SQL, MySQL), MongoDB, Databricks, Unity Catalog, Power BI, Jupyter
Software engineering
Software architecture, Requirements elicitation (story mapping), Scrum, Technical leadership, Git / GitHub
UX
User interviews, Usability testing, Accessibility (WCAG 2.2 AA), Design systems
Research
Systematic literature reviews, Technical writing
Data governance & privacy
LGPD-compliant data masking and handling, AI governance guidelines

Education

  • B.Sc. in Software Engineering, University of Brasília (UnB) · Expected Dec 2028
  • Languages: Portuguese (native), English (fluent)