| Special Seminar Talks

Name of the Speaker: Mr. Kiran Thekumparampil
Venue: ESB-234 (Malaviya Hall)
Date/Time: 04th December 2025 (Tuesday), 11:00 AM
Title: Building AI Agents That Can Operate a Web Browser

Abstract :

We explore recent progress in building AI agents that can operate a web browser using keyboard and mouse actions. These browser-using agents (BUAs) enable end-to-end automation of online workflows such as research and data extraction, form-filling, and shopping. We begin by introducing the pre-trained Vision-Language Models (VLMs) that serve as the primary backbone for BUA systems. Next, we discuss how post-training techniques such as supervised fine-tuning (SFT) and reinforcement learning (RLHF/RLVR) can be applied to adapt the VLMs for agentic browser use. Finally, we highlight some technical challenges faced during training, action execution, and evaluation of such BUAs.

Speaker Bio:

Kiran Thekumparampil (IITM, EE, BTech 2014) is a researcher at the Amazon AGI Labs, where he builds computer-use AI agents. He previously built the current generation of Amazon product search rankers that serve over 300 million customers worldwide. His research spans questions from machine learning, optimization algorithms, and their intersection. Kiran received his master's and PhD in Machine Learning from the University of Illinois Urbana-Champaign and his bachelor's degree from the Indian Institute of Technology Madras. He has also been a part-time researcher at Google and a visiting scholar at the University of Washington, Seattle. He currently serves as an Area Chair for the ICML, ICLR, and NeurIPS conferences.