pvsdevPT

About me

I'm a software engineer with 4+ years building AI-powered products, internal tools and automations in SaaS. I ship end to end, from schema to deploy, and my focus today is AI in production: agents, RAG and evals.Campinas, Brazil · Remote · Open to full-time roles and consulting.

Experience

  1. Software Engineer (AI)

    Dec 2025 – Aug 2026

    • Built internal AI tools (agent skills, benchmarks and evaluation suites) adopted across engineering teams, cutting median PR review time by 30%.
    • Migrated features from a legacy desktop client (WPF, 15+ years of code) to the Angular web platform, improving performance by 70%+ on the migrated screens.
    • Optimized SQL Server queries on the main platform and used agents with observability access to investigate critical production incidents.

  2. Software Engineer

    Apr 2024 – Dec 2025 (junior until Mar 2025)

    • Architected a multi-agent system with 5 specialized agents (OpenAI Responses API) for conversational service and 24/7 support.
    • Replaced a third-party vendor with an in-house AI document analysis pipeline, cutting costs by 85%.
    • Redesigned the customer intake platform with 20+ features, AI suggestions and Stripe payments: +20% conversion.
    • Set up LLM observability with Datadog and an evaluation system for the agents in production.
    • Built the multi-step sign-up with Stripe and Salesforce sync through queues and events.
    • Brought AI tools (Cursor, Claude Code, CodeRabbit) to the team and onboarded 7+ engineers, including senior hires.

  3. Intern and Junior Developer

    Jan 2022 – Feb 2024

    • Built full-stack applications with React, React Native, Node.js, GraphQL and PostgreSQL, from concept to production.
    • Automated testing with Jest, Karate and Detox (E2E), cutting defect detection time from days to hours.

How I work

Measure before trusting
Evals come from real interactions, and production observability shows what tests missed.
The right model for the job
Cost and latency matter as much as quality. Searching documents doesn't need the most expensive model.
The hard part is the harness
The real challenge is in the rules and context of each case, not the prompt.

Education

B.S. Computer Science
UNIVESP · 2024 – present
Electrical Engineering
UNICAMP · 2020 – 2023

Outside work

Path of Exile, web novels, manhwa and manga. See hobbies

Back to home