Data Scientist

Attribute detection in images for fashion retail using machine learning algorithms

• Technologies: AWS S3, Amazon SageMaker, AWS EC2, Google Cloud Platform, GCP Storage, Cloud run, Cloud functions, Docker, TensorFlow, fast API

Managed a team of twelve professionals, responsible for the fraud prevention strategy, models and study environment administration for banking and e-commerce transactions:

• Weekly Fraud Committee presentations with the main strategies and results;
• Daily monitoring of fraud KPIs;
• Daily monitoring and studying of new fraud heuristics performance;
• Build, Deploy and maintain Machine Learning Models for post-anomaly detection for the temporary detention of money in case of fraud;
• Build, Deploy and maintain machine learning predictive models for real-time fraud risk scoring;
• Maintain a useful Data Science environment as well as a Data Lake on AWS for studies and predictive model building;
• Daily monitoring of attacks in order to create rules to stop fraud in E-commerce, Face-to-Face Payments and Pagbank transactions.

Led a team of five Data (Scientist/Engineers) Analysts that:

• Rebuilt the "Online Game Payments Fraud Scoring", reducing losses related to frauds on a average of 30%;
• Built a predictive model for the new Payment API, marketed to online marketplaces;
• Built and deploy predictive models using various machine-learning tools for post anomaly detection for the temporary detention of money in case of fraud, recovering over 50% of a fraud chargeback;
• Test new data sources, measuring the costs (in case of being external) and benefits of new information;
• Study and create new features to improve real-time fraud scoring;
• Set up and maintain a Data Science environment and a Data Lake on AWS.

• Migrate the entire Payment Workflow to a new environment, obeying its limitations and maintaining the previous results;
• Build and deploy predictive models using various machine learning tools for real-time fraud prevention on e-commerce payments;
• Design models to detect anomaly in face-to-face payments, allowing retention and recovery of chargebacks related to crimes;
• Explain complex modelling in an understandable and relatable way;
• A/B Tests to create heuristics to prevent new frauds;
• Credit Scoring Modeling for P2P Loans.