I work on privacy-preserving machine learning: differential privacy, federated optimization, and methods that keep models useful while protecting the data they train on. This fall I start an MS in Privacy Engineering at Carnegie Mellon.
Differentially private second-order federated optimization with regularized Fisher information and Rényi-DP accounting.
NL → JSON → Fusion 360 Python via RAG + LLM (Qwen 2.5). Built at Philips R&I.
VAE-GAN decoder reconstructing visual stimuli from fMRI BOLD signals, with FSL preprocessing.
Policy-gradient actor-critic with max-entropy deterministic annealing over continuous action spaces.
Risk-scoring via BERT embeddings, TF-IDF and autoencoder residuals to flag resume inconsistencies.
Formula-Student vehicle simulation — suspension A-arm forces & bolt safety factors across dynamic events.