AI/ML engineer and tech lead building agents, LLM (Large Language Model) infrastructure and RAG (Retrieval-Augmented Generation) pipelines, on top of two decades shipping production software. I've been taking things apart since I was a kid soldering circuits at age 10, long before my first professional line of code back in 2004.

I lead a development team in Curitiba, Brazil, where I coordinate the architecture and make the technical calls on how we build an AI-powered ecosystem that integrates marketplace, ERP and CRM. It runs in production on .NET, Laravel, React/Next.js (TypeScript), AWS and Python.

Agents, LLM infra & RAG: orchestration, retrieval pipelines and the unglamorous plumbing that makes language models actually useful in production: evaluation, guardrails, cost control, and knowing when not to use an LLM.

Deep learning for medical signal processing (research): training models for electrocardiogram (ECG) diagnosis. Beyond the common diagnostic classes, the current frontier is detecting STEMI (ST-Elevation Myocardial Infarction) and OMI (Occlusion Myocardial Infarction), the cases where minutes decide outcomes and the classic criteria fail the most. That's the what; the how stays in the lab for now. News soon.

CNN1D + Mamba (SSM, State Space Model) for images: exploring sequence-first architectures applied to image data: treating images as 1D scans and letting selective state spaces do the spatial reasoning that convolutions and attention usually monopolize.

SimpleMapper: a zero-configuration object mapper for .NET. Compiled Expression Trees, thread-safe cache, convention over configuration. Born from a simple frustration: mapping code is the most boring code we write, so it should require the least effort. Battle-tested in a full production codebase, not just benchmarks.

• Convention over configuration. Every knob you expose is a decision you're outsourcing to someone with less context than you.
• Boring in production, curious everywhere else. The stack that pays the bills should be predictable; the home lab is where the weird ideas go.
• Depth transfers. Understanding how something works one layer below where you operate is the highest-leverage skill I know.

Self-taught in most things I do: electronics, mechanics, digital painting, and multiple musical instruments. Give me something broken (an engine, an amplifier, a model that won't converge) and I'm happy. I'm convinced the habit of learning things from zero is the actual job skill; everything else is syntax.

Curitiba, PR, Brazil