Currently working on the development of an interpretability system for large language models used in customer-facing Gen-AI applications and regulatory compliance models.
Data Science Modeler
Generative AI Team
Designed the data map to pull the DCC Root Cause flags/notes from the Snowflake environment and performed exploratory data analysis (EDA) to understand the distribution of the various root causes.
Replicated the UDAAP regulation compliance model using the GPT2 frameworks to detect agents’ adherence to regulations during customer interactions.
Supported the development of the No Contact model, leveraging the LLaMa framework.
Anti-Money Laundering (AML) Team
Developed a hybrid machine learning model – NLP (RoBERTa) and Non-NLP (XGBoost) for fraud detection and bias control.
Automated the data preparation workflow in SAS, optimizing the ETL in Snowflake and improving efficiency.
Validated the text preprocessing pipeline using RoBERTa and implemented bias control mechanisms, which includes the removal of sensitive business domain information from texts to mitigate model bias.
Implemented Principal Component Analysis (PCA) to transform unbalanced numeric data samples and utilized SMOTE to balance oversampled data representation successfully, and Information value to weight the features.
