Ekta Samani
Date:
Speaker
Ekta Samani is a Postdoctoral Research Associate at the Robotics Institute, Carnegie Mellon University. Previously, she was a Postdoctoral Scientist at Amazon Robotics. She received her Ph.D. in Mechanical Engineering from the University of Washington (UW), Seattle. Her current research investigates principled approaches to characterizing the reliability of general-purpose AI models within decision-making systems, with a focus on manufacturing applications. She was selected as a 2025 Rising Star in Mechanical Engineering and a 2023 RSS Pioneer. Her honors include the UW Graduate School’s 2024 Distinguished Dissertation Award and the Women Engineers Rise Outstanding Student Award. She serves as an Associate Editor for IEEE ICRA and IEEE RA-L and received a 2026 Outstanding Associate Editor Award for her service to the latter.
Speaker Links: Website
Abstract
General-purpose AI models offer a new source of knowledge for decision-making, but their guidance is unreliable. I will first present our work on LLM- guided optimization for FDM 3D print configuration selection, where an LLM provides guidance within a numerical optimization loop rather than directly selecting solutions. I will then use this setting to motivate an important question: can a decision-making system improve performance by changing its reliance on AI guidance as evidence accumulates? Similar strategies are well established in Kalman filtering, sensor fusion, and ensemble learning, where information from multiple sources is weighted according to estimates of their reliability. I will discuss ongoing work exploring how observed outcomes can inform reliance on AI guidance, and how reliability information can be incorporated into decision-making systems. I will close with a broader perspective on what it means to evaluate the reliability of general-purpose AI models embedded within larger decision-making systems.