HYWE is a research project and experimental computational environment where structured intent metamorphoses into spatial configurations through deterministic design computation.
Hygrid Woven Ensemble
Actively evolving. WebGPU browser recommended.
Important
HYWE is strictly a conceptual layout engine, not a comprehensive architectural modeling or lifecycle management platform. It does not attempt to replicate the detailed parametric modeling, structural documentation, or construction-level detailing inherent to industry-standard building modeling suites. HYWE operates exclusively at the early, exploratory stages of design, focusing entirely on flow-based spatial topology and deterministic volume generation. Evaluating HYWE's outputs against fully-featured, production-ready modeling platforms misrepresents its core intent and operational scope.
HYWE is an investigation into spatial reasoning as a function of computational logic. It explores a methodology where hierarchical programming and relational flow drive spatial generation.
At its core is the Hygrid, an integer hybrid lattice designed for deterministic topology-driven spatial synthesis. By trading geometric isotropy for arithmetic determinism, Hygrid maintains strict integer closure and enables HYWE to execute kernel-free within WebAssembly.
In this system:
- Connectivity is the declared relational graph - the architectural requirement that nodes belong to structured flows and programmatic branches.
- Adjacency is not prescribed as an arbitrary pairwise matrix; it is an emergent geometric and topological consequence of connectivity, directional sequence rules, and lattice packing constraints.
By replacing manual geometric drafting with a visual node tree interface, HYWE lowers the barrier to topological exploration, allowing designers to map out programmatic hierarchies while the engine deterministically weaves this intent into a cohesive spatial Ensemble.
Hywe.mp4
Click any interface preview to view in full resolution.
HYWE functions as an experimental apparatus bridging abstract programmatic intent and physical constraints. The engine translates architectural programming (hierarchical trees, room capacities, and sequence operators) directly into resolved planar and volumetric layouts.
The spatial logic incorporates boundary confinement, enabling configurations to adapt to irregular site boundaries and non-standard footprints. Vertically, this reasoning extends to resolve programmatic stacking, spatial nesting, and multi-level circulation across building massing.
HYWE separates user-facing architectural concepts from internal computational modules:
| Architectural Stage | Domain Concept | Implementation Component | Computational Transformation | Output Representation |
|---|---|---|---|---|
| Intent & Scope | Programmatic Hierarchy & Boundary Rules | NodeTree & BoundaryEditor |
User input graph construction & polygon boundary capture | Abstract Design Intent |
| Encoding | Symbolic Rule Representation | HYWE Syntax | Compact, deterministic serialization | Canonical .hyw String |
| Parsing | Programmatic Flow Tokenization | Lexel |
AST parsing, token extraction, and hierarchy validation | TreeNode Tree |
| Quantization | Lattice Coordinate Allocation | Hexel |
Spatial discretization on the discrete integer Hygrid | Integer Coordinate Lattice |
| Clustering | Emergent Spatial Grouping | Coxel |
Synchronous outward growth and collision resolution | Spatial Cluster Fabric |
| Geometry | Hygrid Spatial Boundaries & Islands | Goxel |
Generates Hygrid-specific geometry, boundary capture, clipping, & islands | Verified Boundary Topology |
| Planar Layout | 2D Spatial Distribution | Xyxel |
Sequence-driven planar placement & orientation sweeps | Planar Layout & SVG |
| Nesting | Sub-space Containment (Xyxel within) | Nexel |
Child cluster nesting within parent coordinate envelopes | Multi-level Hierarchy |
| Massing | 3D Stacking & Volumetric Form (Xyxel above) | Zaxel |
Vertical floor stacking and level elevation assignment | WebGPU 3D Mesh |
| Visualization | Multi-modal Viewports | Graphics |
2D planar SVG rendering and WebGPU 3D massing viewport | Visual Interfaces |
| Evaluation | Spatial Metrics & Verification | Analyze |
Compactness, graph adjacency verification, and PDF export | Performance Reports |
| Exploration | Systematic Permutation | Batch / Teach |
Full-space sequence sweep exploration (24 canonical operators) | Multi-variation Records |
As a downstream capability of its discrete, syntax-first architecture, HYWE provides an automated export pipeline to generate the HYWE Architectural Training Data. Rather than relying on heavyweight or ambiguous 3D file formats, the system pairs natural-language spatial intent with compact, deterministic HYWE syntax strings and Base34 grid encodings.
Design Intent ➔ HYWE Syntax ➔ 24-Operator Sequence Sweeps ➔ Hynteract Ingestion ➔ JSONL Dataset Commit
- Container-First Layout: Each record encodes one container (
L0,L1,N1, etc.) containing all 24 sequence sweep variations positionally aligned with the room header. - Base34 Coordinate Compression: Hexagonal grid coordinates are encoded in Base34 strings for compact representation.
- Full Sequence Sweeps: Every record covers the complete space of 24 canonical sequence rules for each layout container.
The HYWE core spatial engine is dependency-free and executes kernel-free in WebAssembly. The web application leverages Bolero (Blazor on WASM) and Elmish for reactive state management, alongside native WebGPU for 3D rendering.
In this architecture, Syntax is the primary source of truth: every layout is a pure, deterministic transformation of its underlying AST and sequence rules, ensuring identical reproduction across runs on the same engine version.
graph TD
A1[Interactive Node Tree Input] --> B[HYWE Syntax]
A2[Interactive Boundary Editor] --> B[HYWE Syntax]
B --> C(Lexel: Architectural Programming and Flow Parsing)
C --> D(Hexel: Atomic Spatial Primitive)
D --> E(Coxel: Simultaneously Evolving Hexel Clusters)
E --> Gx(Goxel: Hygrid-Specific Geometry, Boundaries & Islands)
Gx --> F(Xyxel: Planar Configuration)
F --> N(Nexel: Spatial Nesting - Xyxel within)
F --> G(Zaxel: Volumetric Stacking - Xyxel above)
F --> Gr[Graphics: 2D SVG & 3D WebGPU]
G --> Gr
F --> H[Analysis: Compactness & Adjacency]
F --> I[Batch Processing: 24 Sweeps]
Gr -.-> ReportLabel((PDF Report))
H -.-> ReportLabel
I -.-> ReportLabel
I -.-> DatasetLabel((Hynteract: AI Dataset))
DesignIntent[Design Intent] --> DatasetLabel
- Language: F# (functional-first design)
- Frontend: Bolero (Blazor on WASM)
- 3D Graphics: WebGPU (native massing)
- HYWE Wiki: In-depth tutorials, spatial concepts, architectural programming guides, and technical references.
- Methodology & Scientific Protocol: Formal definitions of determinism, the adjacency principle, architectural grammar semantics, and benchmark standards.
HYWE is an open project exploring design computation.
Those interested in extending the engine or exploring its procedural logic can refer to the Contributing Guide. Additionally, a technical summary of the architecture is maintained at llms.txt for AI agents and automated analysis.
This project is licensed under the MIT License.








