Research

My research asks one question in seven different settings: how do you build machine learning that respects the physics, uncertainty, and constraints of the real world, instead of treating it as a black box? That question started during my PhD at MIT with Pierre Lermusiaux — stochastic optimal control and uncertainty quantification for ocean path planning, work that produced a US patent and, eventually, grew into the QUEST Lab’s current research program at IISc.

Today that program spans seven threads:

• Physics-Informed ML & Neural Operators — PINNs, neural operators, and diffusion models that bake in physics rather than approximate around it
• Ocean, Weather & Climate Prediction — probabilistic forecasting and climate-model bias correction, the lab’s longest-running thread
• Robotics and Physical Intelligence — optimal path planning under uncertainty, now growing into Physical AI (World Models, VLAs, VLMs, LLMs, and Flow Models), my current primary research focus
• Indian Monsoon & Weather Extremes — onset prediction and city-scale extreme-rainfall forecasting
• Data Assimilation, Bayesian Methods & Uncertainty Quantification — the methodological core underneath the threads above
• AI for Social Good: Ecology & Conservation — agent-based models for human-wildlife conflict and sustainable fisheries
• AI for Education — LLM-based question generation and speaking assessment, aligned to Bloom’s taxonomy and CEFR standards

Each thread is a body of published work — for the full breakdown by pillar, representative papers, and the people doing the work, see Research and Publications on the QUEST Lab site, or the citation record on Google Scholar. For where this research turns into funded, applied projects with industry partners, see Industry Relations.