Summary
Siddartha Devic is a PhD candidate and Graduate Research Assistant at USC specializing in trustworthy machine learning, with six years of research experience across uncertainty quantification, fairness, robustness, and stability of ML systems. His work bridges theoretical foundations and practical evaluation of LLMs, including calibration of reasoning models and post-training techniques to help models better understand their own uncertainty. He has contributed to applied projects at Apple and Amazon on LLM uncertainty and post-processing under distribution shift, and visited Berkeley’s Simons Institute to deepen his theoretical perspective. Earlier research spans networking-aware ML, convex optimization for reinforcement learning, and practical systems like an open-source resilient distributed DNN framework. Based in Los Angeles, Siddartha combines deep theory with hands-on experimentation and a curiosity for agentic post-training methods that reveal subtle failure modes in modern language systems.
6 years of coding experience