Higharc research wins Best Industrial Paper Award at EUVIP 2026

DURHAM, N.C., Oct. 30, 2026 — Higharc, the homebuilding AI company for design through construction, announced that its research paper, "BIM-Native Tokenization for Constraint-Aware Room Layout Synthesis," received the Best Industrial Paper Award at the 14th European Conference on Visual Information Processing (EUVIP 2026) in Luxembourg. Manuel Rodríguez Ladrón de Guevara, Senior Machine Learning Engineer at Higharc, who led this research, accepted the award on behalf of the research team on October 1, 2026.

The paper addresses a fundamental challenge in AI-generated architectural layouts: how to represent building geometry and constraints so a machine learning model can produce usable designs within Building Information Modeling (BIM) environments.

The objective of the research was to automatically arrange furniture, fixtures and casework within existing BIM room models. Achieving that required an AI model to understand room geometry, wall relationships, openings and the placement constraints that govern how spaces function.

"The hardest part was deciding how to represent a room so the model could learn from its geometry and produce something a designer could actually use," said Rodríguez Ladrón de Guevara. "That representation problem, more than the ML model itself, became the central question of our research."

The Higharc team spent approximately a year developing the research results.

The paper introduces the Small Building Model (SBM), a domain-specific Transformer model built around a BIM-native tokenization method. The researchers developed a way to encode rooms as structured sequences of building elements, preserving categorical and continuous attributes such as wall types, dimensions, positions and spatial relationships.

This representation enables the model to analyze and retrieve similar room configurations and generate layouts by predicting the placement of room contents within an existing architectural envelope.

In controlled evaluations, the model demonstrated stronger inventory coverage and navigability than two specialized layout-generation baselines trained on the same data. The results support the researchers' central finding that preserving architectural information in the model's input representation plays an important role in generating usable layouts.

The work also helped inform the research behind Higharc's Generative Building Model (GBM), which powers generative design in the Higharc platform. GBM uses structured building data to generate interior layouts within existing BIM room geometry, accounting for measured walls, doors, windows and the relationships between building elements.

The award-winning paper was authored by Manuel Rodríguez Ladrón de Guevara, Jinmo Rhee, Ardavan Bidgoli, Vaidas Razgaitis and Michael Bergin.

“Housing is one of the most consequential industries in the world, yet it remains largely untouched by modern AI,” said Rodríguez Ladrón de Guevara. “This award recognizes more than a technical achievement. It validates the opportunity to build the foundational infrastructure that makes housing a principal domain for machine learning. We're excited to see this work recognized alongside innovations from across industries, and we hope it inspires more researchers to tackle the enormous challenges and opportunities ahead.”

The research is available at https://manuelladron.github.io/sbm/, with the paper accessible at https://arxiv.org/abs/2512.04832.

‍

About Higharc

Higharc is the homebuilding AI company for design through construction. By generating homes as 3D spatial data, Higharc enables builders and suppliers to better design, estimate, sell and build homes. Higharc works with leading homebuilders and supply chain partners. The company’s rapid rise has earned national recognition, ranking in the top 50 of the Deloitte Technology Fast 500 and Fast Company’s Most Innovative Companies of 2026.

Learn more at https://www.higharc.com.