Machine learning for privacy, accountability, and civil rights.

Interactive simulation framework and graph engine to quantify vehicle anonymity loss across San Francisco's ALPR camera network using time-geographic space-time prisms and negative observation constraints.

A local-first, privacy-preserving ML and OCR pipeline designed for rapid discovery triage across terabytes of sensitive legal records without cloud exposure.

RAG-based classification and retrieval with containerized deployment on GCP; bias-aware stance classification for anti-/pro-trans news content.

Implementation and training of a ~253M parameter discrete diffusion Transformer in PyTorch for non-autoregressive, controllable text generation and fixed-token infilling.

Investigating high-dimensional embedding stability and boundary deformation in 2D UMAP manifolds using differential geometry, alpha-shape concave hulls, periodic splines, and Wasserstein distance.

This research explores whether law enforcement disproportionately patrols certain communities, combining geospatial analysis and communication graphing to uncover patterns. The work supports investigations into potential violations of the Fourth Amendment (Search and Seizure) and the Fourteenth Amendment (Equal Protection Clause).