Jeferson Marques Email & Phone Number

State Of Goiás, Brazil

São Paulo, Brazil

São Paulo, Brasil

- Worked with companies within the Stone Co. group (Stone and Pagar.me) on fraud prevention and receivables anticipation.- Conducted data collection, cleaning, and preparation from various sources, including logs, APIs, bureaus, and internal databases.- Developed identity fraud prevention models, analyzing customer registration data and request logs to identify fraud patterns.- Investigated attack patterns in registrations, building features to highlight fraudulent behaviors.- Created several models, achieving 95% precision and 60% recall in identity fraud detection while maintaining the false positive rate below 5%.- Built pipelines for model training and deployment, integrating with Kubeflow and Vertex Pipelines.- Automated customer monitoring and alerting via Apache Airflow DAGs.- Conducted time series forecasting and anomaly detection to build an MVP for fraud detection.- Reviewed and improved legacy fraud prevention and anticipation models with newer technologies and methodologies. - Technologies: Python, SQL, Pandas, NumPy, Scikit-Learn, Google BigQuery, Storage, Vertex Pipelines, Kubeflow, CatBoost, XGBoost, TensorFlow, Jellyfish, Plotly, PyWavelets, Apache Airflow, GitHub

Goiânia, Goiás, Brasil

- Developed a package and pipeline to update the NLP model.- Analyzed and cleaned public expenditure data and electronic invoices.- Integrated BERT's NLP capabilities with Random Forest, improving classification performance by 5% (Micro F1) and 12% (Macro F1).- The developed model was used to classify public expenditures, identifying over R$ 17 million in misclassified expenses.- Initiated a project to detect overpricing in public procurement. - Technologies: Python, Dremio, SQL, Pandas, Power BI, NumPy, Scikit-Learn, NLTK, Transformers, TensorFlow, PyTorch, Django, GitHub Actions, PyPi