(Jay) Digvijay Wadekar

About Jay Wadekar

I'm an assistant professor in the physics department at UT Austin, in the Weinberg Institute for Theoretical Physics and the Center for Gravitational Physics. My group broadly focuses on developing agentic AI systems for scientific discovery, gravitational-wave astrophysics physics-informed ML, cosmology and particle phenomenology. I also co-chair the NASA AI/ML Science and Technology Interest Group and I collaborate with people in the NSF Cosmic AI institute at UT Austin.

I was previously a postdoctoral fellow at Johns Hopkins University (JHU), and before that a member at the Institute for Advanced Study (IAS). I completed my PhD at the Center for Cosmology and Particle Physics (CCPP) at New York University in 2021, working closely with the CCA cosmology group, and did my bachelors in Engineering Physics at IIT Bombay.

I grew up in Nanded in India. If you don't find me at my office, I may be backpacking, running or practicing latin and ballroom dancing.

Research

Research directions

Verification-based agentic AI for scientific discovery

We are interested in building verification-based agentic AI systems for scientific discovery: building agent systems that have to justify their outputs against data and established physics rather than simply assert them. Current directions include GWAgent for gravitational-wave research, agentic equation discovery, interpretable analytic surrogates for expensive simulations, and benchmarks for evaluating physics reasoning, and multi-agent systems.

Gravitational-wave searches and astrophysics

We developed the IAS-HM pipeline, which searches LIGO and Virgo data using waveform templates that include higher-order harmonics rather than the quadrupole mode alone. Accounting for these harmonics recovers binary black hole mergers that quadrupole-only searches miss, particularly systems with asymmetric masses. Related interests span astrophysical formation channels, waveform modelling, parameter estimation, and population inference.

Machine learning for astrophysics

Applying machine learning to cosmological simulations in ways that yield interpretable, physically meaningful models rather than black boxes. Using IllustrisTNG, we showed how the baryonic content of dark matter halos depends on their local environment, and used symbolic regression to express that dependence as compact analytic equations. Related work on the CAMELS simulations augmented the Sunyaev-Zeldovich flux-mass scaling relation, reducing scatter in galaxy cluster mass estimates and placing strong constraints on baryonic feedback.

Research

Topics

Terms weighted by how often they appear across the titles, abstracts and keywords of my papers.

Code

Software

Updates

News