Sahil Sidheekh

I am an AI researcher working at the intersection of deep generative models, tractable probabilistic inference, and neuro-symbolic AI, with the goal of building systems that are not only accurate, but reliable, interpretable, and trustworthy by design. A central focus of my recent work has been developing tractable generative models — particularly probabilistic circuits — that support exact inference at scale while retaining expressive power, and their applications in diverse domains.

I am especially interested in problems where structure and semantics play a critical role, such as learning distributions on manifolds, incorporating logical constraints into learned representations and composing probabilistic and symbolic components to achieve robust, controllable, and human-aligned generalization.

Beyond the technical foundations, I aim to advance trustworthy human-allied AI systems that know what it doesn't know, can reason under uncertainty while remaining transparent, auditable, and aligned with human values, so that we have models that go beyond predictions to explain, justify, and communicate their decisions in ways that humans can meaningfully understand and trust.