Guduru Sai Deepthi

Fraud and Risk Modeling – PayPal

• Designed and deployed a machine learning fraud detection model for unbranded credit cards, achieving over 60% performance lift over the legacy model, and significantly reducing fraudulent transactions and financial losses.
• Productionized the model into a fully automated bi-monthly retraining and scoring pipeline, ensuring continuous adaptation to evolving fraud patterns and data drift.
• Performed deep statistical and root-cause analysis to diagnose the performance decline of a high-impact merchant, providing actionable insights that informed risk strategy and remediation decisions.

Fraud Analytics and Machine Learning Engineer

• Designed and deployed end-to-end fraud detection systems, leveraging autoencoders, PU learning, clustering, graph algorithms, and GNNs, improving fraud detection performance by up to 6x, reducing false positives from 1:60 to 1:10, and analyzing UPI/IMPS datasets ranging from 1M to over 600M records.
• Built scalable data and ML pipelines using Hive, SQL, Airflow, Dagster, Kubeflow, Kafka, Hadoop, Redis, and Scala, enabling real-time and batch inference for over 5 million daily transactions, and improving data processing and retrieval efficiency by approximately 70%.
• Led advanced feature engineering and representation learning, including a slabbing algorithm to handle skewed distributions (reducing overfitting and cutting 80% manual effort), and cosine-similarity-based dimensionality reduction from 88 features to a compact 3-dimensional space.
• Delivered production-grade analytics and ML tools, including a GPT-based chatbot (Hugging Face), to streamline bank communication and data collection
• Real-time monitoring dashboards using Superset, Druid, and Trino to track fraud trends and model performance.