GRUNDRISS
PREMIUM

The Challenge: Generating 5 AI style variations sequentially would block the user interface and create a frustrating experience. A synchronous approach would time out on batch uploads and make the platform feel slow and unreliable.

Our Decision: A priority-based queue system with Supabase Edge Functions at its core. Each style generation job is queued independently and processed asynchronously — with auto-chaining ensuring each completed job triggers the next automatically while the frontend receives real-time progress updates.

The Challenge: Single image uploads and batch uploads require completely different AI approaches. A single prompt strategy would produce poor results for batch reconstruction — misaligning walls, rooms, and spatial relationships across multiple source images.

Our Decision: An intelligent dual prompt system that detects whether the input is a single floor plan or multiple room photos and adjusts AI prompts accordingly. Single uploads preserve exact wall positions while batch uploads analyze spatial relationships for accurate room reconstruction.

The Challenge: Existing platforms delivered one style per order — forcing real estate professionals to place multiple expensive orders and wait days for each variation. This was the core pain point driving professionals away from competitors.

Our Decision: A unified AI processing pipeline that generates all 5 professional styles — Blueprint, Minimalist, Modern, Furnished, and Isometric — from a single upload. Each style runs as an independent job in the queue, delivering a complete style gallery without additional orders or waiting.

AI-generated raster images cannot be edited or used in professional CAD workflows. Real estate agents and architects need editable, exportable files — not static images that require manual redrawing.

Our Decision: Integration with RasterScan API to automatically convert AI-generated floor plans into accurate vector data. The pipeline detects walls, doors, windows, and room boundaries — storing structured data for downstream editing and export in PNG, PDF, SVG, and DXF formats.

The Challenge:A platform handling AI processing, user authentication, file storage, real-time updates, and subscription management needs multiple infrastructure components working in harmony. Managing these separately would add complexity and slow down a 2-month build timeline.

Our Decision: Supabase as the unified backend — combining PostgreSQL database, authentication, file storage, edge functions, and real-time subscriptions in one platform. This dramatically accelerated development while maintaining production-grade reliability and scalability.

The Challenge: The billing requirements were complex — 7-day free trials, Netflix-style plan upgrades that cancel trials immediately, pay-per-generation credits, and self-service billing management. Building this from scratch would consume weeks of engineering time.

Our Decision: Stripe as the billing backbone with comprehensive webhook handling and idempotent event processing. The Stripe Customer Portal handles self-service billing management out of the box — letting us focus engineering effort on the AI platform rather than reinventing payment infrastructure.

The Challenge: Grundriss Premium targets the German real estate market specifically. An English-only platform would immediately alienate the core audience and signal that the product wasn't built with them in mind.

Our Decision: i18next for comprehensive internationalization — enabling full German language support across every interface, label, and user-facing message. The architecture supports additional languages as the platform expands to new European markets.