The starting point here is a practical one: exposed utilities only stay visible for a short window, and once a trench is backfilled the record is whatever photos and video a crew happened to take. The maker wants to add a 3D visual reference to that record, and picked a dug trench beside a concrete pathway and building wall as the test case.
To build it, the maker recorded footage with Meta Glasses, then ran it through the AEC Platform software using a High Quality profile that outputs a point cloud, mesh and Gaussian splat. The original video sits alongside the splat for comparison, and the right-hand panel renders the mesh of the excavation. As the view pans around the trench, the model viewpoint tracks along with it.
The scene itself is a good stress test for texture capture. There are sections of light blue and grey plastic pipe in the trench, orange traffic delineators with reflective stripes on black hexagonal rubber bases, plus cardboard, loose rocks and general debris. A flexible hose or cable runs past sandbags and a black equipment case. Judging from the side-by-side, the model holds up on both spatial accuracy and surface detail across that mix of shapes and materials.
The open question is the software itself. Asked how it differs from photogrammetry tools like Meshroom, the maker did not spell out what makes it unique. Since the appeal here seems to rest on a simple capture-to-model workflow, that comparison matters, and I would like to see it addressed.