James Verbus

James Verbus working on the LUX dark matter detector
Working on detector hardware for the LUX, one of the world's most sensitive dark matter experiments.

I build AI systems for uncertain and adversarial environments: systems where measurement is hard, feedback is noisy, and getting the right answer requires more than an evaluation score.

After nearly a decade at LinkedIn, most recently as a Senior Staff Machine Learning Engineer focused on anti-abuse, trust, and platform integrity, I chose to leave to focus full-time on frontier AI systems: agents, simulation, evaluation, measurement, and research workflows.

Before LinkedIn, I earned my Ph.D. in physics at Brown working on LUX, one of the world's most sensitive dark-matter detectors. Across physics, platform integrity, and AI, the through-line has been the same: extract weak signals, measure what matters, understand uncertainty, and build systems that remain reliable when the ground truth is difficult to see.

Research

Finding AI-Generated Faces in the Wild

Research and engineering work on detecting AI-generated profile images in real-world settings, including CVPR workshop publication and LinkedIn Engineering write-up.

Ph.D. Thesis

Calibrating the LUX Dark Matter Experiment

An absolute calibration of sub-keV nuclear recoils in the LUX detector using neutron scattering kinematics, improving low-mass WIMP sensitivity sevenfold.

Projects

Isolation forest project logo

Open Source

Isolation Forest

A distributed Spark/Scala isolation forest library for unsupervised anomaly detection, built at LinkedIn and open sourced. Recent major updates added Extended Isolation Forest and ONNX support.

Research / Publications

Talks / Videos

Workshop video thumbnail for LLM and RAG workshop

2025

Brown AI Winter School 2025: Exploring LLMs and RAG

A 2.5 hour interactive workshop at the 2025 AI Winter School, hosted by the Center for the Fundamental Physics of the Universe at Brown University.

Venue/Host: AI Winter School 2025, Center for the Fundamental Physics of the Universe at Brown University.

James Verbus presenting at @Scale

2019

Fighting Abuse @Scale: Preventing Abuse Using Unsupervised Learning

Detection of abusive activity on a large social network is an adversarial challenge with quickly evolving behavior patterns and imperfect ground truth labels. These characteristics limit the use of supervised learning techniques, but they can be overcome using unsupervised methods. To address these challenges, we created a Scala/Spark implementation of the isolation forest unsupervised outlier detection algorithm.

Venue/Host: @Scale Conference.
github.com/linkedin/isolation-forest · Original @Scale page

Open to Conversations

I enjoy comparing notes with people building or studying AI systems. I'm especially interested in the intersection of physics and AI, from world models to reinforcement learning and simulation, and in measurement, adversarial problems, and security. If that overlaps with what you're working on, I'd be glad to hear from you.

Start a conversation