Questions and current work

Research

I work on exact solutions of Einstein's equations, develop models of black holes in cosmological media, study the interplay between cosmic expansion and local gravity, and investigate perturbations and superradiance in dirty black holes.

Black holes in evolving backgrounds

Real black holes are surrounded by cosmic environments that can shape what we observe. These environments can include radiation, matter flows, evolving cosmological backgrounds, and fields with properties associated with dark energy.

My colleagues and I work within general relativity, building and analyzing exact solutions for black holes in these environments. We study static, radially varying anisotropic media, including fluids and networks of topological defects. We also study evolving cosmological backgrounds modeled by dust, radiation, or both. To construct these models, we use generalized Lemaître-Tolman-Bondi (LTB) spacetimes.

Within an effective medium framework, we create baseline models that serve as starting points for adding cosmic structure, testing perturbations, and connecting analytical results with numerical simulations.

Current work includes horizon structure and quasinormal mode spectra of charged black holes embedded in matter with anisotropic pressure, and isotropic coordinates for generalized Schwarzschild solutions.

Inhomogeneous cosmological models

Can the universe appear accelerated without dark energy, once we relax homogeneity and geodesic flow assumptions?

The discovery that the universe is expanding at an accelerating rate led to the idea of dark energy. Understanding this acceleration remains a central question in cosmology.

Together with my colleagues, I study this question using Stephani cosmologies. These are smooth universes without rotation, filled with a perfect fluid whose pressure changes from place to place. In these inhomogeneous models, cosmological acceleration can be reproduced without introducing dark energy.

We use these models to examine how assumptions about matter and geometry affect observations. We develop results such as a closed form redshift-magnitude relation. We continue to test the models against observations of supernovae, baryon acoustic oscillations, and structure growth to see which versions remain consistent with the universe we observe.

Agent systems and human feedback

Semantic dynamics in autonomous agent populations

When autonomous agents share an environment, communicate, and adapt, collective patterns emerge and keep evolving as interaction continues.

We study semantic dynamics in such populations using lexical statistics and embedding geometry. We trace how language shifts, how shared representations form, how memory persists across time, and how collective organization changes. Our work combines longitudinal observation of existing agent communities with controlled experiments that isolate the mechanisms through which semantic structure forms at the population level.

Our goal is to understand when a population begins to drift, stabilize, or reorganize, and whether these changes can be detected before they become obvious in individual outputs.

Rater state bias in learning from human feedback

Human feedback is one of the main signals through which AI systems learn human preferences. Each rating comes from a person whose judgment can change with time, fatigue, mood, and external conditions.

We study how systematic shifts in rater state enter preference data, shape reward models, and propagate into policy optimization. We focus on shifts that accumulate over time or arise from conditions many raters share.

Our goal is to understand how small, structured perturbations in human judgment can become persistent directions in the learned behavior of AI systems.

Related work

Complete publication list