
Martin Piala is co-founder of Blurry, a London-based startup building a workspace for reality capture data: point clouds, Gaussian splats, images and videos in one place, in the browser. Blurry started a few months before 3DISE, and INTERGEO 2026 was its first large trade show, straight in as an exhibitor.
We talk about why Blurry stopped being only a Gaussian splat viewer after seeing that 95% of its users still work with point clouds, why it partners with Teleport by Varjo for processing instead of building everything itself, and why splats in VR headsets are a can of worms it hasn't opened yet. Then AI: why Blurry hasn't shipped a half-baked AI feature, why Martin sees the data format, not the model size, as the limiting factor, and whether there is enough 3D training data. He also shares the funding picture: pre-seed from two VC funds, around three years of runway, and a seed round in the millions planned within six to nine months.
Recorded live at INTERGEO 2026 in the 3DISE podcast studio.
Hosts: Michal Gula and Tomáš Barnáš. Guest: Martin Piala, Blurry.
Chapters:
0:00 Blurry: a Google Drive for reality capture data
0:35 Why Gaussian splatting alone is not enough
1:15 From NeRFs to Gaussian splatting
2:05 Viewing, storing and now processing
2:36 95% of users still work with point clouds
3:59 Can one company do processing and viewing?
4:58 Own processing for 360 cameras
5:40 No hardware: partnering with manufacturers
6:01 Gaussian splats in VR headsets
7:13 Why there's no half-baked AI yet
8:14 Can AI spot BIM vs scan discrepancies?
9:39 Frontier AI models in a small startup
10:39 AI doesn't understand point clouds
11:44 Is there enough 3D training data?
12:35 Is a point cloud just text?
14:33 Pivot or course correction
15:12 Pre-seed funding and the next round
16:11 London HQ and a seed round in the millions
17:11 Wrap-up
Fler avsnitt från "3DISE conference"



Missa inte ett avsnitt av “3DISE conference” och prenumerera på det i GetPodcast-appen.








