• Collaborated on predicting fiducial points on CT images and correcting image distortions for a minimally invasive robotic spine surgery product, under the guidance of a senior scientist in the imaging team. • Developed and integrated a deep learning model that enhanced the software pipeline, reducing RPE error by 15%, resulting in all points passing with less than a 2mm error during algorithm pipeline testing. • Optimized the model for constant-time retrieval, reducing the average retrieval time from 100 ms to 10 ms, resulting in significant performance improvements. Co-authored a research paper based on this study as part of a team.

• Developed a chatbot with an Attention-Based Transformer model based on the tailored use case. • Designed web scraping methods for query-reply format data and integrated open-source datasets (Kaggle) into the training pipeline to overcome challenges with dataset collection. Attained a validation loss of 0.2, exhibiting benchmark-level comprehension of context and responses from the model.