• implemented an algorithm for improving ground truth annotations of a synthetic eye dataset; could potentially improve the real-world dataset annotations from the baseline RANSAC algorithm

• Started creating the creator’s data funnel (AARRR metrics) for FB branded content

Tutored and graded for Natural Language Processing, Computability and Logic, and Principles of Computer Science

• Added a toxicity dataset and the AUC (area under the ROC curve) metric to Parlai, which has 9.5k stars

• Used internal tooling to automatically generate safety reports using the team’s safety bench

• Designed external tooling to automatically generate a model card based on the model file and create additional sections for the model’s creator to fill out.

• Tooling is still in use a year later for creating Blenderbot 3’s model card. (yay!)