My name is Pavel Dolin. I’m a 4th-year CS PhD student and Dean's Fellow at Arizona State University, advised by Visar Berisha and Gautam Dasarathy. I apply statistics and ML theory to guide practical advances in AI/ML algorithms and architectures, with a current focus on reliability, scaling laws, and synthetic data generation. My previous publications include post-deployment monitoring for AI-based digital health (NeurIPS 2025), LLMs’ reasoning about actions and effects (ICLR 2025), and exploration of LLM safety and over-defensiveness (ACL 2024).

Outside of academia, I contributed to researching and architecting an LLM routing pipeline for an AI startup Aisera as an AI/ML research intern in the summer 2025. Before the PhD program, I also spent several years as an ML Engineer at an NLP startup Spiketrap, now part of Reddit.

Prior to Spiketrap, I obtained a BS in Physics from UC Santa Barbara, where I also conducted research in computational materials science, predicting material properties with statistical and ML methods, culminating in a journal publication.