Teach
Data collection in HYWE is user-driven through the interactive Teach tab. The goal of this interface is to pair the designer's exact logical setup with a high-quality natural language description of the resulting space, contributing to the HYWE architectural training dataset.
When designing a layout, the user categorizes their programmatic intent using descriptive tags directly in the interface:
- Author*: The contributor's name.
- Exploration Description*: The designer's idea or spatial exploration intent (minimum 3 words required to prevent generic repetitions).
- Typology*: Residential, Commercial, Institutional (Compulsory).
- Scale: Layout, Building, Masterplan
- Flow: Sequential, Radial, Hierarchical
- Ambience: Organic, Structured, Intimate (Allows for custom descriptors).
- Stage: Ideation, Zoning, Massing
Contributors are encouraged to articulate their unique design nuances, circulation logic, or spatial experience in their own words (either typed or dictated via built-in Voice Transcription).
On clicking Commit to Dataset, HYWE automatically formats the record's description with both the designer's personal narrative and the automated structural breakdown:
- Described: Contains the user's custom spatial description (or remains empty if none was entered).
-
Generated: Traverses the
LevelsandNestshierarchy, circulation dependencies (e.g.Level 1 (Ascending from Elevator Shaft on Level 0)), sub-nests, and tags into an objective structural summary.
This guarantees that every dataset record maintains an informative, schema-compliant description while prioritizing authentic designer insights.
When a user clicks "Commit to Dataset," the engine generates a 2D vector SVG thumbnail, extracts the 24 pre-computed topological variations from the local cache, and fires the structured record to the Hynteract serverless endpoint.
(For a deep dive into the strict JSONL schema and dataset philosophy, see the Data-Collection core guide).