Curriculum Vitae

B.S. Statistics · B.S. Mathematics · B.A. Computer Science · B.S. Data Science
Cumulative GPA 3.86 / 4.00

With A. Jannesari. arXiv:2611.08902, November 2026.

Paper Code, data, registrations Project page

Research project, Advanced Topics in High-Performance Computing (COMS 6250), Department of Computer Science, Iowa State University. Advisor: Prof. Ali Jannesari, SwAPP Lab.

• Ran five registered paired tests on Iowa State's Nova HPC cluster through SLURM, on two serving engines (SGLang, vLLM), two GPUs (NVIDIA A100, H200) and two 7–8B models: a 200-session burst left the engine at zero hits to the end in all 30 burst runs (registered rule met at 9 of 10 rates), every matched no-burst run stayed good, and a client-side session cap prevented the trap in all 30 capped runs. In further registered tests the cap rescued 9 of 9 established traps at the same load without shedding a session.
• Proved that, along FIFO orbits, the cycles of an idealized fluid server on which every turn misses obey a linear renewal recursion whose Perron root, above the all-miss capacity, carries every trapped state to a known self-similar regime; with nothing fitted to trap data, the model matches 57 stuck runs retrospectively. Built the discrete-event simulator, the real-engine load generator and the SLURM job scripts, with each registered test committed to git before its jobs were submitted and every number in the paper traced to its source file and commit or SLURM job. Preprint on arXiv, November 2026; code, data and registrations are public.

Research Collaborator, Department of Mathematics, Iowa State University. Advisor: Prof. David Herzog. Funded by NSF award DMS-2246491.

• Built an end-to-end Python simulation pipeline implementing MALA and projected Euler–Maruyama samplers for overdamped Langevin dynamics, staged from a one-dimensional periodic potential to a solvated ion in a three-dimensional Lennard-Jones bath; estimated self-diffusion and mobility coefficients from long-run trajectory data.
• Derived the closed-form Lifson–Jackson diffusivity D = 1/I₀(1)² ≈ 0.62386 as an exact benchmark and validated the estimators against it before scaling to three dimensions, where the pipeline reproduces published solvated-ion results.
• Author weekly technical reports, including a companion note supporting the group's CLT-based error analysis.

Undergraduate Research Fellow, Department of Mathematics, Iowa State University. Advisor: Prof. Ruoyu Wu. Dean's High Impact Undergraduate Research Award.

• Developing importance-sampling estimators for rare events in mean-field join-the-shortest-queue load-balancing systems, with the change of measure guided by large-deviation asymptotics for the system's empirical measure, and studying how the estimators behave as the number of servers grows.
• Built a reproducible Python simulation framework with deterministic seeding, checked against crude Monte Carlo wherever crude Monte Carlo is reliable. Results are held back pending submission. The project is paused while Prof. Wu is on sabbatical.

Undergraduate Researcher, Department of Mathematics, Iowa State University. Advisor: Prof. Jue Yan. Funded by the NSF D4 NRT. Joint with Mengyuan Yang.

• Adapting von Neumann stability analysis to neural-network PDE solvers: deriving the linearized amplification factor of the learned time-stepping scheme and characterizing how perturbations propagate through it, toward stability conditions on the trained weights analogous to those of classical finite-difference methods.
• Reimplemented the group's solver in Python to enable systematic stability experiments across schemes and step sizes.

Research project, Advanced Topics in Machine Learning (COMS 6730), Department of Computer Science, Iowa State University. Advisor: Prof. Hongyang Gao.

• A forecast of a cumulative quantity is scored only when its horizon ends, yet a miss can become certain before then. Developing a model-agnostic calibration method for quantile forecasts that uses this early, one-sided feedback, now in the experimentation and development phase.
• Built a leakage-audited data pipeline with a sealed test set, a simulation study with pre-registered hypotheses, and delayed-feedback conformal and classical volatility baselines, including a temporal convolutional network. Results are held back until publication.

Research project, Advanced Topics in Artificial Intelligence (COMS 6720), Department of Computer Science, Iowa State University. Advisor: Prof. Yang Li.

• When an agent stalls, when is restarting it worth more than waiting, and how can the probability of a failure too rare for direct simulation be estimated? Now in the experimentation and development phase, starting from two pre-registered pilot designs on real agents, one on restart policies and one on rare failures. Results are held back until publication.

With D. Herzog, O. Smith and D. Wu. Draft in advisor review.

With M. Yang and J. Yan. In preparation.

With R. Wu. Paused during the advisor's sabbatical.

PyTorch and CUDA for neural-network and GPU-accelerated simulation; Python (NumPy, SciPy, pandas, scikit-learn, matplotlib, asyncio); SLURM on GPU clusters and LLM serving engines (SGLang, vLLM); R; LaTeX; git.

Measure-theoretic probability and stochastic differential equations; MCMC (MALA, Metropolis–Hastings) and importance sampling; large-deviation asymptotics; numerical analysis of time-stepping schemes and von Neumann stability; discrete-event simulation and queueing models; online conformal calibration; reproducible computation with recorded seeds, parameter sidecars and pre-registered predictions, so every figure regenerates from saved results.