The Atelier

Inspiration

Managing physical garments in flat spreadsheets is fundamentally broken. Independent creators work with tactile elements like drape, fabric, and spatial location, yet traditional software forces this reality into sterile rows and columns. To mirror a designer's true workspace, the standard dashboard was replaced with a browser-based 3D spatial interface. The room itself becomes the inventory and order management system.

How it was built

End-to-end physical workflows were directly mapped to virtual stations:

The Fabric Wall: Manages raw inventory and low-stock reorder drafts.
Aurora: Handles active garment design and material reservation.
The Pattern Table: Converts scheduled jobs into finished goods.
The Shipping Bay: Processes open orders and generates shipping labels.

Reserving materials on Aurora instantly validates yardage, deducts stock, logs costs, and schedules production. Sealing an order generates a demo label and logs an immutable activity event.

Every visual action triggers a validated API command mutating a rigorous SQLite database, driven by a dependency-free Python backend. The spatial frontend was built entirely with JavaScript, WebGL, and Three.js.

AI Strategy: Artificial intelligence was kept entirely out of the runtime to guarantee deterministic reliability. The live application makes zero AI API calls. Instead, AI served as a powerful build-time engine. Codex accelerated transaction models, handled Three.js interactions, ran accessibility audits, and generated tests. GPT-5.6 compiled raw studio facts into an offline, inspectable Studio Recommendation artifact known as the Sol Capsule.

Challenges faced

Treating a 3D environment as a strict business application rather than a video game required absolute synchronization between WebGL interactions and the underlying SQLite database. Visual updates could not be faked; every spatial movement had to pass strict backend transaction rules.

Furthermore, mapping 3D coordinates back to a flat, logical DOM was essential. This ensured keyboard, voice, gesture, and WebGL-fallback paths remained fully accessible, allowing seamless operation via screen readers without losing operational context.

What was learned

Spatial computing on the web is entirely viable for complex B2B operations when backed by a robust state manager. Most importantly, it was proven that AI-powered projects do not require brittle runtime LLM calls. Leveraging Codex and GPT-5.6 strictly for build-time acceleration yields immense development velocity while preserving the rock-solid reliability that operations software demands.