Interactive GeoAI Urban Layout Generator

I need a small web app that lets me draw or click out a polygon on an OpenLayers base map, type a natural-language prompt (“Redesign this riverside strip into a lively pedestrian zone with mixed-use towers”), and instantly receive a conceptual 3D layout design that respects the selected geometry. The end goal is to experiment with urban-planning ideas in-browser, so the result must come back georeferenced and be viewable as an overlay or downloadable file.

Here is the core flow I have in mind: the map interface captures the polygon, the backend turns that geometry into a clean GeoJSON mask, feeds it—together with my prompt—into an AI model, then returns a layout design image or vector that drops neatly back onto the map in its correct location. Python is preferred for the spatial plumbing (Shapely, GeoPandas, rasterio, etc.) and for orchestrating calls to the chosen generative model. Whether you lean on a large-language model with diffusion, a vector-based generator, or another service is open to discussion as long as it produces an easy-to-read site plan.

Deliverables (minimum viable):
• OpenLayers front end with draw/select tools and prompt input
• Python backend API that receives geometry + text, performs spatial preprocessing, calls AI, and returns the design
• Map overlay of the generated layout with basic opacity toggle and download to GeoJSON/PNG
• Brief setup guide so I can run everything locally

Acceptance criteria:
– A 500 m × 500 m test area renders within ~30 s
– Output aligns correctly with the original map tiles when re-projected
– No hard-coded API keys; configuration via env variables

If you already have GeoAI or prompt-engineering prototypes, I’d love to see them when you reply.
Thnaks!