Sripad Karne

Sripad Karne

Doing independent research in mechanistic interpretability, AI safety, and Physical AI, as well as supervised edge inference research at Columbia with Prof. Kostis Kaffes.

Previously, I worked on multi-agent LLM optimization (AgentOpt) at DAPLab with Prof. Tianyi Peng and Dr. Wenyue Hua, and was an AI Engineer Intern at IBM.

I'm a Master's student in Data Science at Columbia University, graduating December 2026. I did my undergrad at UC San Diego, studying Cognitive Science & ML, advised by Prof. Zhuowen Tu.

Blog  ·  Personal  ·  CV

Publications

How Far Do Auto-Interpretation Labels Generalize? A Controlled Study Across Languages, Scripts, and Rewordings Code
Sripad Karne
AACL-IJCNLP 2026 (main conference)
What the Guard Misses, the Robot Executes: Implied Harm in VLA Instructions Code
Sripad Karne and Arjun Balaji
Science of Physical AI Safety (SPAIS) Workshop @ CoRL 2026
Recall Is Not Protection: Evaluating Safety Monitors Against Model Compliance Code
Sripad Karne
JUDGe Workshop @ NeurIPS 2026 (poster)
Which Model for Which Role? Cost-Aware Pareto Search over LLM Agent Configurations
Qian Xie, Wenyue Hua, Yueli He, Sripad Karne, Armaan Agrawal, Nikos Pagonas, Eugene Wu, Kostis Kaffes, Nairen Cao, and Tianyi Peng
Resource-Aware Agentic AI (RAAAI) Workshop @ NeurIPS 2026
One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations
Sripad Karne
Unifying Concept Representation Learning (UCRL) Workshop @ ICLR 2026
Machine Learning–Predicted Risk Trajectories for Incident Chronic Kidney Disease and Associations with Post-CKD Outcomes
Aaron Boussina, Amy M. Sitapati, Soo-Young Yoon, Sripad Karne, Jongwoo Seo, Woo-jung Kim, and Hyeon Seok Hwang
Journal of Medical Systems, 50(1):86, 2026

Preprints

Safety Monitors Mostly Catch What the Model Already Refuses Code
Sripad Karne
Under review at ICLR 2027 (main track); extended version of the JUDGe workshop paper
Which Model for Which Role? Cost-Efficient Pareto Recommendation for LLM Agentic Workflows
Qian Xie, Wenyue Hua, Yueli He, Sripad Karne, Armaan Agrawal, Nikos Pagonas, Eugene Wu, Kostis Kaffes, Tianyi Peng, and Nairen Cao
Under review at ICLR 2027 (main track); extended version of the RAAAI workshop paper
AgentOpt v0.1 Technical Report: Client-Side Optimization for LLM-Based Agent Code
Wenyue Hua, Sripad Karne, Qian Xie, Armaan Agrawal, Nikos Pagonas, Kostis Kaffes, and Tianyi Peng
Technical report, arXiv 2026
A framework-agnostic optimization layer for client-side model selection in multi-agent LLM systems, released as an installable package.

Experience

Independent ResearcherJan 2026 – Present
AI Safety, Evaluations & Mechanistic Interpretability
Cross-lingual SAE interpretability, safety monitor evaluation, and safety for robot foundation models.
Graduate Researcher, ML SystemsSep 2026 – Present
Columbia University · Advised by Prof. Kostis Kaffes
Optimizing LLM inference for edge hardware.
AI Engineer InternMay 2026 – Aug 2026
IBM
Graduate Researcher, Agentic AI SystemsJan 2026 – May 2026
DAPLab, Columbia University · Advised by Prof. Tianyi Peng & Dr. Wenyue Hua
Optimizing agentic AI infrastructure.
Biomedical AI InternMay 2025 – Oct 2025
UC San Diego Health · Advised by Prof. Aaron Boussina

Grants & Funding

Awards & Honors

Contact

Email: sk5695 [at] columbia [dot] edu
LinkedIn · Google Scholar · GitHub · X