I built an open-source PyTorch toolkit for model interpretability, implementing SHAP, Integrated Gradients, and feature attribution across tabular and vision data. I also analyzed neural network activation patterns to diagnose failure modes and attribution drift.
For a tabular foundation model project, I benchmarked deep learning architectures against baselines and developed Bayesian ensemble methods to estimate uncertainty and risk. I’ve also participated in AI and software hackathons, including an individual Qualcomm Snapdragon® AI Lab challenge.
