Launching today

Decomind — AI for real-state agents

Launching today

AI for real estate agents. Try the 3D home module free

1 follower

AI for real estate agents. Try the 3D home module free

1 follower

Visit website

Real estate

Decomind helps real estate agencies win and sell listings with AI: valuation, listing copy, virtual staging, photo retouching, lead follow-up. Its newest module, built with GPT-6 Astra, turns a floor plan into a walkable 3D home. Used daily by agencies in Spain; developed in Madrid at the Lanzadera accelerator. Try it free.

Decomind — AI for real-state agents gallery image

Free

Launch tags:Design Tools•Artificial Intelligence•3D Modeling

Launch Team / Built With

OpenAI

GPT-6 Astra Challenge

What real task does your product handle with GPT-6 Astra?

Maker

GPT-6 Astra reads a real-estate floor plan and its furnished top-down render and returns the full 3D model as typed JSON through the Responses API: every wall with its openings, every room with its polygon and type, floor and wall finishes per room, and every piece of furniture with position, footprint, orientation, colour and material. Working on a 0-1000 grid over the image with structured output is what makes it reliable: in metres the model hallucinates, in image space it is precise. Pure Python turns that into metres and Three.js builds a walkable home in the browser. Astra also handles corrections in natural language: 'the sofa is dark green and faces the window' re-reads the render with the current model and returns the fixed furniture, finishes and room types without regenerating the image. Real-estate agents use it to show a listing before it is furnished or even built. Try it without signup at app.decomind.es/try-3d

Report

4h ago

Maker

📌

Hi Product Hunt! I'm Manuel Rodríguez, co-founder of Decomind. Decomind is an AI toolkit for real-estate agents in Spain. It covers the whole listing cycle: property diagnosis, valuation from official data (notaries, cadastre, ministry of transport), listing copy for the portals, virtual staging, photo enhancement, a branded listing report and lead follow-up. Agencies use it every day to win listings and publish them the same day. For the GPT-6 Astra Challenge we built our newest module: turn a floor plan into a walkable 3D home. Upload a plan, and in about four minutes you get a furnished top-down view and a 3D home you can walk through and share by link. How it works under the hood: gpt-image-2.5 draws a furnished top-down render from the plan. GPT-6 Astra reads the plan and the render and returns walls, openings, rooms, finishes and every piece of furniture (type, position, footprint, colour, material) as typed JSON through the Responses API. Working in image space with structured output was the key: in metres the model hallucinates, on a 0-1000 grid over the plan it is precise. Pure Python turns that into metres (scale from the plan's dimensions, snapping, corner merging) and Three.js builds the home in the browser with real textures, physical sky and a first-person walk. You can then fix the result with a sentence ("the sofa faces the window", "the kitchen is the top-left room") and Astra re-reads the render with your correction. The code was written with Codex and verified with Claude Code; the runtime runs on Astra. Two weeks from first commit to what you can try today. What works well: clean plans with dimensions, furniture placement, finishes. What still bites: L-shaped sofas become straight, hand-drawn plans confuse the scale, and complex plans with roofs or light wells sometimes get read as terraces. Try it free at app.decomind.es/try-3d (no signup). I'd love your feedback, and difficult floor plans are very welcome.

Report

4d ago

OpenAI

GPT-6 Astra Challenge

What real task does your product handle with GPT-6 Astra?

Maker

Report

4h ago

Maker

📌

Report

4d ago