vitor pereira
I design and deploy agents that execute real workflows inside the tools your company already uses — with logs, rules, human approval, and documentation.
in production
how it works
A message, an email, a file, or an order.
It reads, extracts what matters, and checks it against the process rules.
The ones your company already uses. No need to switch systems.
Anything that can't be undone waits for a person to decide.
You can see what was done, when, and why.
example
An agent that does the work — not one that only answers questions.
Integrations and routines that run on their own across your company's tools.
An assistant that answers about each customer's data, inside the system.
Each agent only reaches what it needs, and the rule lives on the server (not in the prompt text).
2025 · Own product · Co-founder & CTO
AI-powered recurring billing for insurance brokers — from file to reconciliation.
2025 · Co-founder · Full-stack co-lead
Specialized Search and Rescue (SAR) technology for aviation and defense.
2026 · Pixel Educação · Software Engineer
AI agents that work for your business 24/7, no technical team required.
2024–2026 · My Group Metrics · Principal Engineer
In-product analytics and AI for WhatsApp communities at scale.
evidence before claims
These are not polished demos. They cover failures, boundaries, and decisions that appear only after a system starts operating for real.
operational recovery
Why storing the original event made recovery and replay part of the architecture instead of an emergency patch.
security
A penetration test found cross-customer access. What changed in permissions and adversarial tests.
agent architecture
How tool design reduces ambiguity, blocks invalid actions, and makes agent behavior easier to verify.
No. The agent takes over the repetitive part of one bounded workflow; a person stays accountable for the outcome and approves high-impact actions. If the goal is to replace a whole team on day one, this is not the right place to start.
The agent only gets the permissions it needs, fails visibly, and asks for a human decision when an action is irreversible or outside its contract. Its actions are logged, so you can see what happened and fix it.
No. The agent works inside the tools your company already uses, as long as they can be accessed through an API.
A pilot for one workflow takes 21 to 30 days, from R$ 20,000. The exact number comes out of the scoping session, which is included. The first conversation, the AI Automation Diagnostic (30 min), is free.
The boundary lives on the server, not in the prompt: the agent only reaches the data allowed for that workflow, and high-impact actions go through human approval and are logged.
Show me the current flow, the tools involved, and where work gets stuck. We start by deciding whether an agent helps or merely adds complexity.
Free, no strings attached.
We use cookies to understand how the site is used and to improve your experience. You can accept or reject. Privacy policy