AWS Annapurna Labs

As a data scientist in the Educational Platforms team at Tecnologico de Monterrey, I was in charge of designing and developing ML models for improving executive decision making as well as modeling user demand, behavior and churn.

As a Machine Learning Research Intern in ASAPP.inc's ML Research Team (linked below), I worked on stochastically combining pre-training for two production LLMs through data augmentation. The clustering and next-utterance-prediction LLMs in production were trained using different context windows (clustering: past + future, next-utterance-prediction: past only) and by stochastically varying these context windows, we could train both models as one. My work showed promising results as it allowed for the reduction of training costs by 2 while incurring a minimal performance hit on either task.

AWS.

pytorch.

pytorch-lightning.

object oriented programming.

experimental design.

Self-supervised learning.

NLP.

https://www.asapp.com/resources/papers