Research Assistant
CurrentWorked with Travelers on developing a transformer model to classify insurance claims data for worker’s compensation estimation.Utilized Python libraries such as Numpy, Huggingface (Transformers & Datasets), and Pandas.Explored different data cleaning approaches on claims data to reduce noise while retaining essential information, using measures such as TF-IDF and feature attribution in the transformer model.Finetuned BERT model for a classification problem on AWS EC2 instances; Initial models achieved an accuracy of ~90% on evaluation sets.