I grew up in a small town pulling apart Pentium 4 beige desktop towers in Bhusawal, India, convinced that the right piece of silicon could unlock new worlds. When I finally saved for a GPU in 2009 (NVIDIA 9400 GT), games became experiments: sliders turned into differential equations, frame rates hinted at dynamics, and I drifted toward pure math just to keep up with my own questions about motion and intelligence.
That restlessness pulled me into interdisciplinary labs, where I translated biological physics into software. I began by building chemical computing agents and CUDA-accelerated reaction-kinetics simulators, then explored in-silico DNA computers that implemented AI-like algorithms. During my PhD, I focused on scaling numerical methods and active-matter/living fluids solvers that matched high-resolution experimental observations and hinted why some biological cells adopt spherical shapes. Later, at Harvard CSE-Lab, I shipped GPU-heavy reinforcement learning tooling and continued publishing on living-fluid physics, because building systems kept teaching me more than describing them.
Today I am exploring a question that sits underneath all of that work: how do we collect data from living systems without disturbing them? Many of our best biological measurements are destructive snapshots. A sample is fixed, lysed, or sequenced once, and the trajectory that produced it is lost. But biology is dynamics, and the things I care about most, how a state unfolds, when it shifts, and what drives it, live in time series rather than single frames. So I am building systems to non-invasively read enough signal to recover the dynamics underneath. This work began with initial support from the Revival Fund.