Saurav Panigrahi

I work across AI systems, research engineering, and AI safety. My work has ranged from database autotuning with reinforcement learning and LLM post-training and alignment at Zoho Labs to AI safety research with Robert McCarthy at UCL and Lionel Levine at Cornell University, along with contributions to training and evaluation for high-stakes domains like medical reasoning at MEDARC.

Works

  1. Report cover icon Technical Report: Side Effects of Character Training: Quantifying Cross Constitution Drift in LLMs Research report on behavioral spillover from character training.
  2. Report cover icon Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks Benchmark suite for evaluating medical reasoning across diverse clinical tasks.
  3. Report cover icon Technical Report: Investigating Intrinsic Self-Preservation in LLMs Research report on when models resist shutdown, redirection, or control.