HYWE – Hygrid Woven Ensemble
About HYWE
HYWE - Hygrid Woven Ensemble
HYWE is a zero-dependency, browser-native computational design sandbox where structured intent resolves into spatial configurations through deterministic topological logic. HYWE and the HYWE Architectural Training Data repository form an active, deterministic data collection ecosystem.
For more detailed information regarding the core engine, design computation logic, and layout features, refer directly to the HYWE GitHub Repository README.
Ecosystem Components & Sync
The HYWE ecosystem operates as a closed-loop data curation and design pipeline across three distinct components:
-
HYWE Engine (Core & Client Sandbox): The F#-driven client interface compiling to WebAssembly via Bolero. Resolves discrete spatial topologies and generates real-time SVG 2D plans and WebGPU 3D volumes.
Learn more in the HYWE GitHub Repository.
-
Hynteract (Ecosystem Data Layer): The private serverless JavaScript/Node.js API ingestion layer deployed via Vercel. It secures client sessions, maps linguistic architectural prompts, captures Base34 spatial tokens, and formats structured JSON Lines (.jsonl) payloads.
-
HYWE Architectural Training Data (ML Dataset): The public repository hosting structured synthetic architectural layouts currently in its active data collection phase. This token-efficient dataset will provide the foundation for training custom, layout-aware AI models.
Access the database, schema definition, and licensing on the Hugging Face Datacard.
The 8-Stage Core Pipeline
The HYWE computational engine processes designer intents through eight deterministic compilation stages:
| Stage |
Component |
Logic |
Output |
| 1. Intent |
Interactive Node Tree & Boundary Editor |
Defining spatial relationships and site bounds. |
Design Intent / Constraints |
| 2. Encoding |
HYWE Syntax |
Compact, deterministic Base34 alphanumeric serialization. |
.hyw State String |
| 3. Parsing |
Lexel |
Flow parsing and architectural program extraction. |
TreeNode Hierarchy |
| 4. Formation |
Hexel & Coxel |
Atomic units and simultaneously evolving layout clusters. |
Geometric Fabric |
| 5. Distribution |
Xyxel |
Planar Coxel configuration and 2D spatial layout. |
SVG Plan View |
| 6. Massing |
Zaxel |
Multi-level Xyxel stacking and 3D volumetric extrusion. |
WebGPU Volumetric Model |
| 7. Expansion |
Batch & Teach |
Procedural permutation processing and active data collection. |
Dataset Ingestion (via Hynteract) |
| 8. Insight |
Analyze & Report |
Spatial adjacency metrics and automated layout audit. |
PDF Design Report |
The 6 Core Topological Concepts
-
1. Deterministic Topology: Adjacency resolves strictly through procedural topological rules rather than probabilistic drafting, ensuring math-precise layout repeatability.
-
2. Architectural Programming: Abstract building programs (hierarchical trees, circulation weight, boundary limits) compile directly into resolved geometries.
-
3. Spatial Hierarchy: Recursive, self-similar space partitioning distributes nested child spaces (L1, L2, etc.) dynamically within parent Coxel hosts.
-
4. Base34 Topology Encoding (HYWE Syntax): High-efficiency serialization format that compresses layout relationships into standard text tokens, allowing stateless sharing via URL fragment hashes.
-
5. Integer-Based Partitioning: Custom arithmetic and discrete orthogonal-hexagonal lattices replace floating-point coordinates, preventing browser-specific rounding bugs.
-
6. Semantic Architectural Intent: Declarative natural language design prompts are mapped directly to discrete spatial coordinates via the Hynteract pipeline, building rich dataset cards.
Technical Stack
- Core Language: F# (Functional-first design)
- Web Framework: Bolero (Blazor running client-side WebAssembly)
- Graphics Layer: WebGPU (Native hardware-accelerated 3D massing)
Repository & Ecosystem Links
