At Unico, I developed and monitored machine learning models processing over 2 million transactions a day in real time. I also built streaming analysis pipelines to identify and mitigate data and concept drift.
I reduced model inputs by 50% without performance loss, and migrated legacy feature stores to optimized feature views, cutting operational costs by 60% and speeding up query execution fourfold.
At Topaz, I contributed to online fraud detection products and built an internal BDD automation framework for deterministic validation of machine learning rules.
My PhD research at Federal University of Rio Grande do Norte focused on semi-supervised data stream classification in non-stationary environments. Across my projects, I’ve worked on real-time ML inference, expert ensembles, and model testing pipelines.

