The main theme of my research is the use of Bayesian probability models and sampling-based inference algorithms, and, more recently, in-context learning with Large Language Models, for (a) both generating and recovering structured semantic knowledge (e.g. entities, topics and their relationships) from text and semi-structured data, and (b) using such knowledge for decision making. As part of my dissertation research at University of Maryland, College Park, and during my stint as Assistant Professor at IISc, Bangalore, I extended probabilistic modeling techniques, such as parametric and non-parametric mixture and admixture models, temporal point processes, and clustering processes, to simultaneously accommodate text and different types of syntactic structure, and applied such techniques to a wide variety of domains and tasks, such as de-duplication and data cleaning in Databases, topic discovery, word sense disambiguation and sentiment analysis in Natural Language Processing, and code analysis in Software Engineering.Â
Previously, as a Principal Scientist at TCS Research, I led the Knowledge Based AI Research Group, where I applied my research to build AI assistants for knowledge workers. I led the development of AI tools for Technical Interviewing and Assessment and Data Transformation and Migration that were integrated into TCS products and also won TCS and Tata Group innovation awards. I also led an exploratory research collaboration with Prof. Mausam at IIT Delhi, on Robust, Low-resource Question Answering over Knowledge Bases.
Even earlier, as a Senior Research Scientist at IBM Research India, I was part of the research team developing IBM Debater, an AI agent that learns to engage in persuasive argument with a human expert. I also led the development of a tool for Social Influence Analysis, which was deployed for the 2016 Wimbledon Championships to track top online influencers, and a tool for Voice of Customer Analytics (VOCA), which was piloted successfully in customer environments across multiple domains.