About Me

I'm a PhD candidate in the Physics Department at the University of Michigan, working in theoretical cosmology under Prof. Dragan Huterer and Prof. Camille Avestruz. My research focuses on dark energy, modified gravity, and large-scale structure, combining statistical methods and observational data to test extensions of standard cosmology.

Alongside my PhD work, I am also deeply interested in machine learning, optimization, and efficient AI systems. I have worked on related problems in LLM quantization and numerical optimization, including collaboration with PolyLabs Inc. on solver design and model efficiency. I am especially drawn to problems where mathematical structure leads to practical algorithmic and systems improvements.

Skills

Programming

Python  ·  C/C++  ·  SQL  ·  Bash  ·  Fortran  ·  CUDA (via PyTorch)

ML & Data

PyTorch  ·  PyTorch Geometric  ·  scikit-learn  ·  NumPy  ·  SciPy  ·  pandas  ·  Matplotlib

Methods

Machine learning  ·  Graph neural networks (GNN)  ·  Statistical modeling  ·  Numerical optimization  ·  Bayesian inference  ·  MCMC  ·  Uncertainty quantification  ·  Hypothesis testing  ·  Experiment design

Systems & Infrastructure

Linux  ·  Git  ·  Slurm  ·  HPC  ·  Parallel computing  ·  GPU workflows (CUDA / Apple MPS)

Communication

Delivered technical talks in 1,000+ member international collaborations. Designed and taught a university-level summer course translating complex quantitative concepts for non-specialist audiences.

Projects

LLM Quantization & Optimization Solver

PolyLabs Inc.  ·  Mar 2026 – Present

Built evaluation pipelines for LLM compression and developed batched PyTorch solvers for binary least-squares problems — including both exact (sphere decoding) and heuristic (ICM) methods. Ran systematic experiments across N = 1K to 1M instances on CPU, CUDA, and Apple MPS, analyzing convergence, stability, and quality–runtime tradeoffs. Found that restart budget dominates over sweep count for harder matrices, enabling more efficient parallelism. Results contributed to a 14× speedup over the baseline in their production pipeline.

Stack: Python · PyTorch · CUDA · Apple MPS · NumPy

Graph Neural Networks for 3D Simulation Data

Built PyTorch Geometric pipelines for graph-based learning on 3D simulation data, including full data preprocessing, model training, and evaluation under noise and distribution shift. Implemented and compared GAT, GCN, and GIN architectures, evaluating predictive performance and robustness across different graph representations and perturbation settings.

Stack: Python · PyTorch · PyTorch Geometric

Bayesian Inference & Forecasting Pipeline

Built scalable Python pipelines for large-scale data processing, Bayesian inference, and high-dimensional statistical modeling on observational datasets in an HPC environment. Designed and executed end-to-end analysis workflows for weak-signal detection in noisy data, including baseline comparison, robustness checks, ablation-style validation, and systematic bias quantification. Also developed a tomographic cross-correlation forecasting pipeline — quantifying performance sensitivity to event count, distance error, and localization accuracy in a high-dimensional inference setting.

Stack: Python · NumPy · SciPy · Slurm · HPC

Academic Research

My academic work is in theoretical and observational cosmology, with a focus on dark energy, modified gravity, and large-scale structure. I am a member of the DESI collaboration. Full publication list: InspireHEP  ·  Google Scholar.
5,000+ citations  ·  h-index 16  ·  21 papers (18 published)

Modified Gravity Constraints with DESI

Derived constraints on Horndeski and non-minimally coupled gravity models using DESI DR2 and Y1 data. Helped build and run the EFT of Dark Energy analysis pipeline, explored scale dependence in full-shape clustering, and contributed to cosmological parameter estimation.

Modified gravity constraints

J. Pan and G. Ye, Non-minimally Coupled Gravity Constraints from DESI DR2 Data (2025), arXiv:2503.19898.
M. Ishak, Pan, J.*, R. Calderon et al., Modified gravity constraints from full shape modeling of clustering measurements from DESI 2024, J. Cosmology Astropart. Phys. 09 (2025) 053, arXiv:2411.12026.

Hubble Constant from Galaxy–GW Cross-Correlations

Developed a tomographic cross-correlation pipeline using galaxy surveys and gravitational-wave events to constrain H₀ without host-galaxy identification. Performed Fisher forecasts and quantified uncertainty scaling with event count and localization errors.

Hubble constant GW

J. Pan, D. Huterer, C. Avestruz et al., Determining the Hubble Constant through Cross-Correlation of Galaxies and Gravitational Waves (2025), arXiv:2510.19931, under revision in Phys. Rev. D.

Initial Conditions for Horndeski Gravity

Derived consistent adiabatic initial conditions for Horndeski gravity within the EFT framework and implemented them in cosmological Boltzmann codes, resolving long-standing inconsistencies in early-time evolution.

J. Pan, M.-X. Lin, G. Ye, M. Raveri and A. Silvestri, Consistent Initial Conditions for Early Modified Gravity in Effective Field Theory, Phys. Rev. D 112 (2025) 083551, arXiv:2506.17411.

BAO Analysis in Modified-Gravity Models

Studied the robustness of the standard BAO analysis in Horndeski models, finding negligible bias in BAO compression when applied to modified-gravity data.

BAO contours

J. Pan, D. Huterer, F. Andrade-Oliveira and C. Avestruz, Compressed baryon acoustic oscillation analysis is robust to modified-gravity models, J. Cosmology Astropart. Phys. (2024) 051 [2312.05177].