We learn from our alumni in this interaction series, often technically, sometimes semi-technically.

Chaithanya Bandi NUS, Singapore

Chaithanya Bandi is a faculty in the Department of Analytics and Operations at NUS Business School. Prior to this, he was a faculty at Kellogg School of Management. He obtained his PhD in Operations Research from MIT after graduating from IIT Madras with a BTech in Computer Science. He is broadly interested in the problems of decision making under uncertainty, incomplete information and risk with applications to operations management. In particular, he is focused on developing Robust Optimization based models to formulate key problems in applications such as queueing control, risk optimization, mechanism design, and online algorithms; with applications ranging from e-commerce, healthcare, crowdsourcing, data-centers, and cloud-computing.

The advent of generative AI and foundational models has opened up promising avenues to enhance the efficiency of many service systems. However, their wider application has been stymied by the persistent presence of model errors and inaccuracies. These errors represent an intrinsic characteristic of current AI models, and not just a transient hurdle. In this talk, we present a Robust Optimization based approach that embraces this reality and seeks to optimize performance within this context. We present a Robust optimization-based algorithmic framework and develop the Fast Algorithm for Classification and Truth-validation (FACT), which aims to address the critical need for rigorous truth-validation in service systems. FACT is designed to leverage the decision-making competencies of both humans and foundational Large Language Models (LLMs) through a dynamic task routing strategy, informed by task complexity, projected cost, and the capacity of the decision-making entity. We present our results from a pilot implementation at a customer service portal of an online education company, which demonstrates significant efficiency benefits. This approach offers a promising trajectory for future research and practical implementations, especially in the context of systems where AI and human decision-making must be effectively integrated.

Organizers