Riiid AIEd Challenge 2020

Created models for knowledge tracing application i.e. the modeling of student knowledge over time. 100M samples for 400K users were provided by organizers. Created LightGBM and Transformer models utilizing Pytorch and Google Cloud, AWS. Ranked top 4%.

Tweet Sentiment Extraction

Given a tweet and a sentiment (positive or negative), the objective here was to pick out part of the tweet (word(s) or phrase(s)) that reflects the sentiment. Utilized Roberta model from Huggingface (Pytorch) library for this Natural language processing task.

Google Quest Q&A Labeling

Created models for improving automated understanding of complex question answer content. This was a regression task with 30 labels. Utilized multiple Natural language processing models with Huggingface (Pytorch) library. Utilized BERT, Roberta model from Huggingface (Pytorch) library for this Natural language processing task. Ranked top 3%.

Jigsaw Unintended Bias in Toxicity Classification

Challenge in this task was to build models that recognize toxicity and minimize unintended bias with respect to mentions of given identities. Created an ensemble of NLP models - RNN, BERT Base, BERT Large and GPT 2. Ranked top 5%.