Evidence over enthusiasm
I test a claim about AI against a live process, a control and an audit before I repeat it.
About
From enterprise systems engineering to Enterprise AI leadership and doctoral research.
How the work developed, what it consists of now, and the principles I kept.
Background
I started as a software engineer, building the enterprise systems companies run their daily operations on. Correctness, integration and availability were the standard from the beginning, and they still are.
I then spent much of my career on enterprise architecture and software delivery for large organizations, largely on SAP systems. In those environments correctness is audited, integration is unavoidable and change is expensive. Working inside those constraints is how I learned to evaluate new technology.
I moved from building systems to leading the people who build them: first a consulting practice of 10, then a technology organization of 50, then delivery programs of up to 80. Program leadership taught me that most technology failures are execution failures: unclear ownership, unmanaged dependencies, and no plan for adoption.
In 2021 I joined NTT DATA to lead Data & AI delivery, and since 2023 I have been Director of Data & AI, responsible for enterprise AI strategy, governance and portfolio with a team of 15. I lead Guepardo AI, the AI portfolio of an enterprise platform used by more than 550 customers, spanning predictive analytics, deep learning, generative AI and AI agents.
I also represent the Brazilian operation in NTT DATA's Global Innovation Team, within an organization present in more than 60 countries. Brazil has unusually demanding tax and fiscal requirements, which makes it a strict test of whether an AI system holds up in production.
In parallel I went back to study. After a degree in Computer Engineering, I took a specialization in applied AI at PUCPR, then an MSc on peak prediction in time series for smart grids, and since 2024 I have been a PhD candidate there. My thesis is on adaptive algorithms with privacy guarantees. I teach at PUCPR and have been an associate member of the Brazilian Computer Society since 2024.
The two halves feed each other. Research shows me what these methods can do; delivery shows me what an organization can operate.
How I work
Execution is the hard part
The model is rarely the constraint. Ownership, data access, adoption and operational fit usually are, and those get decided in meetings, not in architecture.
Privacy as a design input
In regulated domains, data protection changes what can be built at all. I set it at the start of a project, alongside latency and cost.
Build for the long term
Enterprise systems stay in service for years. I build so they can still be understood, audited and changed long after the original team has moved on.
Working domains
The areas where my professional and academic work overlap.
- Enterprise AI strategy and governance
- Generative AI and AI agents in business processes
- Data & AI leadership
- Privacy-preserving machine learning
- Federated learning
- Homomorphic encryption
- Data stream mining and concept drift
- Finance, accounting and tax processes
- Large-scale enterprise systems delivery