Production AI for companies

AI that works on Monday morning

Every company has seen an impressive AI demo. Few have seen one working on Monday morning, with real data, impatient users and a mistake that costs money. The difference between the two is not the model; it is the engineering around it: clean data, the right context, continuous evaluation, clear limits and a path to a human when the model does not know.

That is what I build. The most concrete proof is Cravamos, a news portal where eleven specialised AI agents publish every day under human editorial responsibility, with identified sources and review. It is not an experiment: it is an operation.

What I do

Diagnosis of where AI creates real value in your operation, and where it does not. Assistants and agents connected to your data (RAG) with access control. Process automation with human review at the points that matter. Classification, extraction and summarisation of documents at scale. Language-model integration into existing systems, with predictable cost. Evaluation (evals), guardrails, monitoring and audit trail.

How I decide what to use

By the right question: which error is acceptable? If none, AI does not decide alone; it suggests and a human approves. If some, we measure the rate and define the limit. Model, vendor and architecture come after that answer, never before. And sensitive data stays where the law and common sense say it should.

For whom

Companies in Cascavel, Toledo and Foz do Iguaçu that want to move from pilot to operation. Cooperatives and industry with document volume and repetitive processes. Portals and media. Health and research, the field where I am building ELUCENIA. Technology teams that need someone who has operated AI in production, not only read about it.