• Achieved the biggest sale in the squad's history, reaching 9x the corporate target stipulated for the year in the total contract, by applying an algorithm developed in Python that combines machine learning models based on performance and cost, in a proof of concept with a of the 10 largest banks in Brazil.• Developed R&D projects aimed at advancing the state of the art in the techniques used by data scientists.• Led strategic meetings to share knowledge on knowledge about advanced techniques widely used in machine learning, including model interpretability in the internal audit process of large banks and feature engineering strategies.

• Contributed to the development of internal libraries, accelerating the study and experimentation process of the company's data scientists.• Participated in the design of solutions for banks and financial institutions, addressing challenges throughout the credit cycle.

• Participated in the development of a solution for transmission tower inspection, utilizing drones for field image capture, focusing on the classification of anomalies and defects, significantly reducing manual effort in image analysis, most of which lacked anomalies.• Deployed the production architecture for the models that composed the solution, containerizing them and serving them on a Kubernetes infrastructure in Azure (AKS), while also setting up a monitoring dashboard using Grafana.

• Conducted research and experimentation with SOTA algorithms in the field of federated learning, adapting frameworks (i.e., Flower, PySyft) to meet the specific needs of the project.• Developed new datasets to support researchers in the field globally.• Published at the international conference VISAPP (Computer Vision Theory and Applications) the work "FedBID and FedDocs: A Dataset and System for Federated Document Analysis".

• Developed an application for capturing and detecting barcodes on automotive production conveyors in the Stellantis group, reducing manual verification error by ~63%.• Used YOLO for object detection and Tesseract as OCR.

• Implemented a system to generate textual analysis metrics, including readability, coherence and cohesion, using NLTK and SpaCy.• Applied these metrics to a textual feedback classifier for university teachers to manage large volumes of individual assessments that achieved 95.1% accuracy on a set of 1000 texts assessed by real teachers, using scikit-learn.