Research focuses on learning-based 3D reconstruction and content synthesis using neural fields and hybrid explicit–implicit geometry, leveraging differentiable and inverse rendering. Builds text/image‑conditioned generation methods and efficient data/annotation pipelines.