Technology and Research Driving Homiwoo’s Growth

At Homiwoo, innovation sometimes takes unexpected paths. Arnaud Guèze, the company’s first CIFRE PhD candidate, is working on a fascinating challenge: reconstructing 3D spaces from partial data. A few photos and a short textual description are enough for an apartment to emerge in three dimensions. This research, led by Damien Rohmer and Marie-Paule Cani from the Computer Science Laboratory of École Polytechnique (LIX), and supervised within Homiwoo by Mathieu Ospici, could transform the way we interact with real estate. Between complex algorithms and practical applications, this project turns the invisible into reality.

“Homiwoo makes it possible to develop essential tools that will transform the way we visualize and interact with real estate spaces.”

An Innovative Approach to 3D Reconstruction

At the heart of Arnaud Guèze’s research lies a radically different philosophy from traditional scientific approaches. “The complexity of our research comes from the fact that we only have sporadic data,” the young researcher explains.

Rather than being an obstacle, this uncertainty becomes a driver of innovation. The team is developing adaptive methods that improve as more data becomes available, while still being capable of producing results even in highly constrained situations. This flexibility is based on a complex mathematical architecture using graph theory to structure and process spatial information in an optimal way. A graph is a mathematical structure made up of points, called “nodes” or “vertices,” connected by lines called “edges” or “links.” In the context of Arnaud Guèze’s research, the graph is used to represent the spatial relationships within an apartment: each room becomes a node, and the connections between them, such as doors and openings, become edges.

The objective is clear: “To take into account all the information available and, from there, infer new suggestions that are highly likely,” summarizes the PhD candidate. It is a major technical challenge that fundamentally rethinks traditional real estate practices, turning every available data point into an entry point for a richer spatial understanding.

AI-powered space creation

To fill in the gaps in the available data, Arnaud Guèze relies on generative artificial intelligence: “These are the same methods used for image generation or video generation,” he explains.

The principle is fascinating: “The AI first learns what an apartment is before becoming capable of creating new spaces,” the researcher explains. But unlike simple replication, the technology demonstrates true controlled creativity. “In fact, AI is capable of creating new forms, new rooms, and doing so according to the constraints we impose on it.”

This approach is not without challenges. Unlike traditional software, where every step is fully understood, artificial intelligence remains largely unpredictable. It is impossible to know precisely how it reaches its conclusions, which makes the researchers’ work significantly more difficult. This is especially true in still largely unexplored fields such as apartment reconstruction, where teams are essentially starting from scratch, without solid references to build on. “It is long-term work that requires unwavering perseverance and countless iterations before achieving the expected results,” says Arnaud Guèze.

A Revolution Underway for the Real Estate of Tomorrow

The practical applications of this research could be revolutionary for the real estate sector. “From simple photos and a brief description, we will be able to estimate what the apartment looks like in space, in 3D,” the researcher explains. This visualization opens up unprecedented possibilities for real-time personalization.

“For example, for a listing of a dated house, we could ask the tool for style proposals in real time,” the researcher imagines. It would also become possible to simply enter a few keywords to visualize the space in a minimalist, contemporary, or rustic style. The AI could change the furniture while preserving the room layout, allowing users to project themselves into the property and perhaps make a purchase they might otherwise not have considered.

In the longer term, the ambitions are even broader. “AI could also make it possible to project oneself into renovations or interior work,” Arnaud Guèze envisions. Artificial intelligence could analyze a simple photograph of an outdoor space and automatically propose complete layouts, precisely calculate the quantities of materials required, and even identify the best offers available on the market. This end-to-end approach would radically transform project planning, making it possible to design, estimate, and source a renovation project in a matter of hours, where previously it would have taken weeks of research and multiple quotes.

Although this work is still at the academic research stage, it reflects Homiwoo’s long-term vision and its ability to anticipate changes in the real estate sector. The ambition behind this research could mark a significant milestone in the evolution of property technologies and open up entirely new prospects for the industry as a whole.