UNFOLD

UNFOLD - Spatial Intelligence Lab What if understanding the physical world were as simple as having a conversation with AI? Today, transforming technical drawings, photographs, and complex ideas into interactive three-dimensional environments requires specialised knowledge, expensive tools, and significant manual effort. I built UNFOLD to challenge that limitation. UNFOLD is an experimental spatial intelligence platform that transforms images and natural-language descriptions into structured, interactive 3D representations. Instead of simply generating visual content, it attempts to understand how objects are composed, how their components relate, and how those relationships can be explored and manipulated. Using Meta Muse Spark 1.3 for multimodal interpretation, I developed a pipeline that identifies candidate components, spatial relationships, geometric properties, and structural constraints. I then process these interpretations through my own deterministic geometry and validation engines, built primarily with TypeScript and Three.js. A fundamental principle of my architecture is that AI predictions should never automatically become verified facts. UNFOLD distinguishes observations, inferences, approximations, and unknown information, incorporating human review before applying relevant changes. Inside the unified workspace, users can explore 3D scenes, inspect individual components, isolate structures, preview modifications, undo changes, and preserve their work. The demonstration features a detailed procedural hydraulic excavator with articulated components, alongside experimental automatic spatial reconstruction. The excavator demonstrates the visual and mechanical sophistication I aim to generalise across different object categories. My ambition extends beyond CAD software or conventional 3D viewers. I envision UNFOLD becoming a new interface between human curiosity and physical understanding, supporting education, engineering, design, and technical exploration. I am building UNFOLD around one idea: the future of AI should not stop at describing the world. It should help people understand and interact with its structure.