Worked on pipelines for modeling patient-therapist conversations:

• Used word/sentence embedding methods (word2vec, transformers) to build representations of conversation language that were used to predict insights from the conversations.

• Built models for predicting patient risk & symptoms from conversations.

• Did various analyses & produced visualizations for scoping out useful insights from conversation data to show our clinicians/members.

Tools used: Pytorch/Spacy/Huggingface (NLP), Spark/SQL/Numpy/Pandas/Sklearn/Matplotlib (General data processing/modeling), Databricks (platform)

At Earnin, I did various projects on the risk team:

• Engineered features, defined labels & tuned hyperparameters of regression/tree-based models for predicting credit risk.

• Did feature engineering & performance analysis to build rule-based fraud model from scratch for new savings feature.

• Built dashboards for monitoring various types of risk.

Tools used: Python/SK-Learn/Pandas/Numpy (developing models), SQL/Redshift/Sisense (analysis/monitoring), AWS (platform)

I worked on PsyNeuLink, a modeling framework for cognitive neuroscience:

• Built an abstraction for parsing cognitive models made in PsyNeuLink and building them out identically in Pytorch, enabling them to be trained faster.

• Improved model training time by ~100x without sacrificing the ease-of-use/modularity PsyNeuLink provides.

• Wrote extensive tests to ensure abstraction correctness and measure training speedup for different model architectures.

Tools used: Pytorch/Numpy