“Hallucination Detox: Sensitive Neuron Dropout (SeND) for Large Language Model Training” accepted to NeurIPS 2024 SafeGenAI
“Distillation of Transformers with Efficient Mechanisms” accepted to ICML 2024 ES FOMO II Workshop; Winner of UofT AI Project X 2024 Competition
Improved Edgebank for temporal link prediction by introducing a novel metric sigma (σ); implemented temporal collaborative filtering using PyTorch and SciPy
Built Moviefy.AI, a deep collaborative filtering movie recommendation system for 1M users, with Docker, CI/CD, Kafka, MLFlow, and Canary deployment
Hackathon winners: (1) Real-time emotion-based music recommendation app using DeepFace and Spotify API; (2) Stroke risk predictor achieving 87.37% accuracy and 79.89% F1 score using Scikit-learn, XGBoost, and PyTorch