• Helped contribute to a BERT-based entity disambiguation/linking system that improves performance on entities that are rare or not seen during training. Worked on adding labels for weak supervision, and using entity embeddings for downstream tasks such as text generation, among other tasks.
  

• Helped develop deep learning models to improve computer vision systems for in-store analytics at RADAR, a NYC-based retail technology company.
• Designed an end-to-end process for training models to perform “person re-identification” in video footage.
• Collected training footage from in-house studio, researched and built person re-identification models, and stress-tested models to identify weaknesses.
  

• Researched deep learning methods for automatically identifying energy infrastructure in satellite imagery. Trained deep learning-based object detection models, including Faster-RCNN, YOLOv2, and RetinaNet.
• Researched model generalizability across different geographies and image resolutions, to inform future work.
• Poster placed 2nd (tied) at the 2019 Duke Research Computing Symposium.
• Paper accepted to the 2020 IEEE International Geoscience and Remote Sensing Symposium (IGARSS).

• Researched and employed predictive analytics at a Silicon Valley innovator of wearable biosensor technology.
• Designed machine learning algorithm that automatically detects abnormal heart sounds from raw, unstructured audio recordings with high accuracy.
• Published conference paper (presented at 2018 IEEE ICMLA conference).
• Algorithm incorporated into PCT patent and non-provisional patent applications (filed September 2019).

Teaching assistant for three departments at Duke University (Statistical Science, Computer Science, Mathematics).

Courses include:

• STA 325: Machine Learning & Data Mining
• STA 523: Statistical Computing (PhD-level)
• CS 101: Intro to Computer Science
• MATH 122L: Calculus II with Applications