Ryan Baten
About Ryan Baten
Ryan Baten is a Lead Data Scientist at Transamerica, where he has worked since 2021. He previously held roles at the University of Colorado Boulder and Transamerica, focusing on data science and system administration.
Work at Transamerica
Ryan Baten has been serving as the Lead Data Scientist at Transamerica since 2021. In this role, he has focused on enhancing customer support through the application of causal inference techniques, which optimized follow-up emails. He previously held the position of Data Scientist at Transamerica from 2018 to 2021, where he developed a system of models for generating call simulations. This system was instrumental in training new agents with 20,000 simulated calls, contributing to improved training processes within the organization.
Education and Expertise
Ryan Baten studied at the University of Colorado Boulder, where he earned both a Bachelor of Science (BS) and a Master of Science (MS) in Computer Science. His academic journey spanned from 2014 to 2018, during which he gained foundational knowledge and skills in data science and computer programming. His education laid the groundwork for his expertise in developing models and implementing advanced data techniques in his professional roles.
Background
Before joining Transamerica, Ryan Baten worked at the University of Colorado Boulder in various capacities. He served as a Linux Student System Administrator from 2015 to 2018, where he managed system operations. Additionally, he worked as a Course Assistant for four months in 2018, supporting faculty and students in the Computer Science department. These roles contributed to his technical skills and experience in academic environments.
Achievements
During his tenure at Transamerica, Ryan Baten led the technical transition of the Data Science team to AWS Sagemaker. This initiative resulted in significant cost savings and improved efficiency in model training and deployment. He also implemented a linear RoFormer model in PyTorch for call disposition, achieving an 8 macro-F1 score improvement over baseline methods. Furthermore, he built proof of concept use cases for large language models, including retrieval-augmented generation with internal knowledge datasets.