• Automated deep learning training dataset construction by implementing a computer vision pipeline (OpenCV) to extract quantitative data from heatmap images.

• Experiemented with ESM-1b transformer and Alphafold-2 based protein representations in PyTorch to predict protein/peptide interactions with a hierarchical statistical model

• Fine-tuned seq2seq transformers (BART, PEGASUS) for clinical note summarization to reduce hallucination after filtering low-quality training notes and applying constrained beam search

Designed a data pipeline in R to gate (filter) and interpret physical characteristics of thousands of individual cells

from a flow cytometer

Assisted in media experiments to compare the expression variation in isogenic cells of S.cerevisiae