My background started from programming languages and software systems, then moved into JAX-based scientific machine learning, NLP, Bayesian modeling.
I am now extending that foundation toward geometric learning, Physical AI, world models, and 3D reconstruction. My current work is reading-heavy because I am still mapping how modern 3D vision, representation learning, probabilistic modeling, and deployable AI systems should be connected.
My GitHub therefore contains both older exploration across languages and systems, and newer work around JAX, world models, 3D vision, and Physical AI.
LeJEPA is one of the bridges between my earlier work in representation learning and my current focus on 3D vision and Physical AI. I read it not merely as an SSL method, but as a possible replacement for heuristic visual backbones such as DINO in geometry-heavy systems like VGGT or DUSt3R-like pipelines.