Pavan Gollapalli

Data Scientist @GM Financial

US

Built real-time pipelines with Kafka and Spark for fraud detection, enabling distributed processing and ingestion of large transactional datasets. Reduced data latency and processing time by 40% through efficient streamlining. Developed deep learning models (Autoencoders) for anomaly detection, fine-tuning architectures to identify subtle fraudulent patterns. Achieved 94% AUC-ROC by leveraging advanced feature extraction and optimizing performance metrics. Integrated LLMs (GPT-4) within a RAG framework for fraud analysis and reporting. Combined transactional data with contextual retrieval using Pinecone, enhancing compliance and decision-making for fraud investigations. Optimized fraud models using XGBoost, Isolation Forest, and Logistic Regression, complemented by SMOTE for imbalanced datasets. Conducted hyperparameter tuning with Optuna, improving recall and reducing false positives significantly. Deployed scalable solutions using AWS Sagemaker and Lambda. Containerized applications with Docker and implemented real-time monitoring via CloudWatch, ensuring robust model performance in fraud detection workflows. Designed visual dashboards using Tableau and Matplotlib to track transaction patterns, fraud trends, and key performance metrics, improving stakeholder insights and operational transparency. Automated model retraining pipelines with Sagemaker, enabling continuous updates to adapt to evolving fraud patterns. Ensured system reliability and high scalability for real-time predictions across environments.