I move between formulation, implementation, and empirical investigation.

My background spans autonomous and interactive systems, multimodal tracking and reasoning, long-context architectures, distributed model training, GPU kernels, and low-latency inference infrastructure. Across those settings, the recurring problem is the same: make a model’s behaviour measurable, then build enough of the stack to test the explanation.

My current public direction centres on reinforcement learning and alignment for embodied AI—particularly the relationship between representations, latent world-action models, policies, and the feedback used to shape them.

I completed graduate research at IIT Kanpur on efficient and robust discriminative manifold learning and multi-modal multi-object tracking. Public technical writing and code are linked from this site; employer-confidential work is intentionally not described here.