Machine learning scientist
I am an ML research scientist and engineer working on online learning, anomaly detection, agent memory and orchestration, and learning problems in stochastic networks. My work includes theoretical research and deployed machine-learning systems.
Industry work
At AWS, I have worked on machine-learning systems for security, agent memory and orchestration, and structured-data applications.
GuardDuty for RDS Protection uses unsupervised online models to learn account-specific behavior and detect anomalous database login activity. I worked on the statistical methods and production system from initial research through launch and subsequent operation.
I have worked on memory, orchestration, evaluation, and personal knowledge graphs for agents. Related capabilities are used in Amazon Quick and the SageMaker assistant.
I have worked on language-model systems for structured data, including text-to-SQL and schema disambiguation. This work contributed to capabilities in Amazon Quick and to the ODIN research project.
Research interests
My research includes online learning and sequential decision-making, anomaly detection, stochastic networks, random graphs, and distributed algorithms.
Collaboration
Background
AWS, 2020–2026. Worked on adaptive security systems, agent memory and orchestration, and structured-data capabilities for Amazon Quick and SageMaker.
Academic training. Postdoctoral researcher at UC Berkeley with Venkat Anantharam; PhD in Computer Science and Mathematics at UT Austin with François Baccelli; B.Tech. and M.Tech. from IIT Madras.