Abstract: Recent advancements in 3D scene representation have prioritized rendering speed at the expense of accurate light transport. In my talk, I will introduce Radiant Foam, a novel approach that bridges this gap by leveraging a volumetric mesh-based differentiable representation. Unlike traditional rasterization-based methods, Radiant Foam enables real-time ray tracing without specialized hardware, offering precise handling of reflections, refractions, and other complex light interactions. Our framework parameterizes a 3D scene as a Voronoi-based volumetric foam, ensuring smooth optimization and adaptive resolution. By integrating efficient ray-tracing techniques originally overlooked in modern vision research, we achieve rendering performance comparable to 3D Gaussian Splatting (3DGS) while retaining the flexibility of NeRF-like models. This method opens up new possibilities for real-time graphics, neural rendering, and photorealistic scene reconstruction. Join me as we explore how Radiant Foam redefines the balance between efficiency and realism in differentiable rendering.