About Me
I work as Head of AI at Sourcegraph in London. Prior to this, I was the Director of ML at Sharechat, and even earlier, I was a Staff Research Scientist & Area Tech Lead at Spotify in London. I have a PhD in Machine Learning from University College London (UCL). I studied Computer Science and Mathematics at BITS Pilani.
My PhD research was focused on developing efficient machine learning models for user tasks & need understanding, knowledge discovery and decision optimization. During my PhD, I co-founded a company UserContext.ai with my supervisor Emine Yilmaz. During my PhD I’ve spent time at Microsoft Research in Redmond (2015, 2016), New York (2016) and London/Cambridge (2015). Prior to joining UCL, I worked briefly at Goldman Sachs (2013).
Research Interests
My research interests revolve around developing scalable learning systems that power multistakeholder marketplace platforms. I work on multi-objective ranking, reinforcement learning, user modelling and large scale evaluation. I am interested in teaching machines to better understand (& thereby model) users & optimize for different stakeholder objectives.
Over the past few years, my research has focused on two key areas:
-
Recommendations in a Marketplace.
Multi-sided platforms facilitate efficient economic interaction between multiple stakeholders among which there are different individuals with assorted needs. While traditional recommender systems focused specifically towards increasing consumer satisfaction, such platforms need to optimize for multiple stakeholder objectives. My recent research has focused on designing search & recommendation frameworks that power such multi-stakeholder platforms. This includes developing multi-objective ranking/recommendation techniques, quantifying stakeholder objectives, developing user understanding modules and developing joint optimization modules. -
Learning from User Interactions.
When users interact with online services (e.g. search engines, recommender systems, conversational agents), they leave behind traces of interaction patterns. The ability to record and interpret user interaction signals and understand user behavior gives online systems a vast treasure trove of insights for improving user engagement and experimentation. I have developed a number of ML methods that learn from such user information, including bayesian non-parametric techniques to infer user intents & tasks, tensor based approaches to learn user models, neural approaches for gauging user satisfaction and developing evaluation metrics.
For short introductions to the above topics, you can check out my recent talks below.