Kaggle AI Competition Platform

Learning Agency Lab Automated Essay Scoring 2.0 Competition

This competition aimed to develop automated essay grading models that provide efficient, timely feedback to teachers, reduce grading costs, and enable broader integration of essays into student learning assessment.

I achieved a top 4% score (108th of 2706 entries) on the efficiency leader board.

Equity in post-HCT Survival Predictions Competition

This competition challenged participants to improve allogeneic Hematopoietic Cell Transplantation (HCT) survival rate predictions, fostering equitable and successful patient outcomes.

I explored experimental machine learning approaches, including using Large Language Models (LLMs) for feature engineering. I developed PyTorch solutions with custom loss functions optimized for competition metrics, and employed Generative Adversarial Networks (GANs) for synthetic feature generation and test-time training, visualized via TensorBoard.

Code available here. I achieved a score in the top 37% of participants.

Google Universal Image Embedding Competition

This competition challenged participants to develop advanced image representations that are effective across diverse object types, departing from traditional domain-specific models. The aim was to retrieve relevant images based on object similarity from a varied dataset across many visual domains.

I achieved a score in the top 44% of participants.

Google Research Football with Manchester City F.C.

This competition, a collaboration between Manchester City F.C. and Google Research, challenged participants to train AI agents to play football against each other. The goal was to develop agents capable of mastering the sport's complex balance of short-term control, learned concepts like passing, and high-level strategy, making strategic decisions from 19 possible actions to outscore the opponent.

I achieved a score in the top 45% of participants.