I'm a data scientist and machine learning enthusiast who is passionate about building real-world solutions, with a strong focus on NLP and multimodal machine learning. In my free time, I also try to stay active on Kaggle by participating in various competitions. Here are a few that have gone well:
-
Optiver Trading at the close: In this competition the task was to predict the US stock closing movements. I used a stock clustering approach for generating new features and purged cross-validation technique for training models that were a part of my final solution that ranked 283rd out of 4436 participants.
-
CommonLit Readability Prize: In this competition the task was to predict the complexity of literary passages for grades 3-12 classroom use. Our final solution included a RoBERTa-Large model with custom attention head and layer-wise learning rate decay that ranked 88th out of 3633 participants.
-
Multilingual Abusive Comment Detection (ShareChat/Moj): Built a transformer-based system for detecting abusive content across low-resource languages and ranked 2nd out of 54 teams.