AI agents · production systems · São Paulo

AI agents that hold up in production.

Twelve years shipping payment, banking and crypto systems that weren’t allowed to fail. I bring that same discipline to AI agents — so they survive contact with real users, real load and real money.

Talk to me Read the CV

Open to: engineering leadership · AI advisory · founding-team conversations

Pix · C6 Bank ~50 devs on my tooling Cryptohub · Mercado Bitcoin

12 yrsshipping to production

Pixpayments, cards, crypto

~50 devsship on my tooling daily

4 repospublic, auditable

What I do

Most agents demo well. Few survive Monday.

The gap between a demo and a system a business depends on is engineering discipline. That is the whole of my career, and now it is what I build agents with.

Agent systems

Multi-LLM orchestration

Conversational agents across multiple models on a multichannel platform: real-time message routing, cost-tiered model selection, and the guardrails that stop an agent from hallucinating halfway through a flow. The hundred-plus agents in rc-project are planners, executors, reviewers, code-security auditors and councils. A taxonomy, not a headcount.

Foundations

Systems that can’t fail

Pix, card issuing, crypto and high-throughput payment APIs. Go, Java, Kotlin and Node on AWS, over Kafka, SQS and Kubernetes. Money moved through code I owned.

End to end

Architecture to on-call

Requirements, solution design, implementation, deploy, production tuning. I stay on the thing until it works where it matters: in front of users.

Proof

Open source, in real use

Not side projects. Tools that a team ships with every day.

rc-project

An agent plugin that drives the full lifecycle of AI-assisted development: idea → PRD → tech spec → tasks → execution → review → remediation. Runs inside Claude Code and other agent hosts. Over a hundred agents live in it — planners, executors, reviewers, code-security auditors, loop pipelines, and councils that argue a decision out before it ships. Cost-tiered specialists route recon to cheap models and hard reasoning to strong ones. Every artifact is plain markdown.

→ Standard engineering tooling for a team of ~50 developers, shipping to enterprise clients in telecom, digital banking and solar energy.

lib-acp

A Node client for Claude Code and Codex over the Agent Client Protocol. Subprocess management, JSON-RPC framing and session lifecycle collapse into three methods: create a client, open a session, stream prompts.

gh-repo-audit

Points at the junk in your public GitHub profile: forks you never touched, projects abandoned years ago, secrets you committed by accident. Prints the gh commands to fix it and executes nothing. I ran it on myself and deleted 77 repositories.

gh-achievements

The GitHub API doesn’t expose profile achievements, but it exposes the numbers behind them. This computes what’s computable and prints the exact unlock criteria for the rest. One shell script, nothing but gh.

Track record

Where the discipline came from

Present · 4apps

Co-founder

AI agents, AI influencers and AI-generated content. Pre-launch.

2025 – present · Escale

Tech Lead, Solution Engineering

Conversational agents running on multiple LLMs across a multichannel platform. I lead the engineering behind new client rollouts.

2024 – 2025 · Foursys

Java Consultant

Technical reference on the company’s strategic microservice projects.

2023 – 2024 · TQI / PagBank

Senior Software Engineer

High-throughput payment APIs in Go, Java and Kotlin.

2022 – 2023 · C6 Bank

Tech Lead

Pix, Payments, Corporate Payroll and Corporate Billing.

2021 – 2022 · Mercado Bitcoin

Senior Engineer, Solution Architect

Credit Card and Cryptohub, at Latin America’s largest crypto exchange.

2020 – 2021 · PagBank

Senior Software Engineer

Payment APIs, on the Tróia squad.

2008 – 2020 · Sucocítrico Cutrale

Systems Analyst → Project Coordinator, Tech Lead

Twelve years learning to ship software a business actually depends on.

Straight answers

The questions you were going to ask

What are you open to right now?

Engineering leadership on agent products, AI advisory for teams moving from demo to production, and founding-team conversations. I’m co-founding 4apps, so anything I take on has to be worth the trade. Tell me what the problem is and I’ll be direct about fit.

Do you actually run agents in production, or just build tooling?

Both, and that’s the point. At Escale I design and ship conversational agents on multiple LLMs and tune them in production. rc-project is the engineering process around that work, hardened until a fifty-person org adopted it.

Why should a bank or fintech trust an AI person?

Because the AI is the recent part. Before it: Pix and payments at C6 Bank, credit cards and Cryptohub at Mercado Bitcoin, transaction APIs at PagBank. I know what an incident costs when real money is on the line, which is exactly why I don’t ship agents that only work in a demo.

Can I see the code?

Yes. Everything above is public on GitHub. Start with rc-project if you want to see how I think about agent architecture, or lib-acp if you want to see how I write a small, boring, correct library.

Where are you, and how do you work?

Brazil (São Paulo), remote-first, working with teams in Portuguese and English. I take ownership end to end: architecture and requirements through implementation, deploy and the on-call that follows.

Next step

Let’s make your agents survive production.

If you’re building with AI agents and need them to hold up under real users, real load and real consequences — write to me. I answer every serious message.

Email me LinkedIn

rodolfo.chicone@4apps.ai