Shobhit is a student at Stanford studying CS and Math. Currently, he splits his time between training long-video understanding models + world models at SAIL (Fei-Fei Li, Ehsan Adeli, Juan Carlos Niebles), building robots in his dorm room, and post-training research at Armadin. At Armadin, he focuses on three key initiatives: scaling training (OPSD, PPO, GRPO, etc.) to push the frontier of cyber capabilities in open-weight models, generate autonomously improving long-horizon synthetic environments, and optimizing serving infrastructure for hyperattacks against large enterprises. On campus, he’s involved with Stanford AI Club and previously helped organize TreeHacks. Outside of work, you can find him speedrunning word games, playing poker, engaging in racquet-based sports, watching a movie, or occasionally hitting legs.
[overview]
[Scholar Journals]