FLOORA: A Human-Aligned Domain-Specific Language Model for Architectural Design

2026-09-28

AI Summary

This study addresses the limited generative capability of general-purpose large language models in structured engineering tasks such as architectural layout generation. To this end, we propose a lightweight, domain-specific language (DSL) model tailored for architectural design. Methodologically, we design a token-efficient DSL paired with a custom tokenizer, and optimize human alignment through supervised fine-tuning (SFT) combined with reinforcement learning incorporating verifiable rewards. Experimental results demonstrate that this compact model substantially outperforms large frontier models, achieving a 96% win rate under vision-language model evaluation, an 89.3% preference rate in human assessment, and significantly improved geometric validity. Overall, this work establishes an efficient, domain-adapted training paradigm for vertical applications.

Abstract

Foundation models are powerful generators, but many engineering domains require structured representations that general-purpose systems handle poorly. We introduce FLOORA (Floor Layout Optimization with RL Alignment), a family of small domain-specific language (DSL) models for architectural layout generation. With specialized data and alignment, our 0.6B model outperforms much larger frontier models, achieving VLM judge win rates up to 92.0% on out-of-distribution real-world buildings and 96.0% on synthetic buildings. Human evaluations further corroborate these results, with FLOORA selected as the best model in 89.3% of evaluations. FLOORA combines a token-efficient DSL, custom tokenization, domain-specific pretraining, supervised fine-tuning (SFT), and reinforcement learning (RL) with learned human-preference and verifiable rewards. This pipeline improves architectural and geometric validity, supported by extensive empirical evaluation and ablation studies. Although focused on architecture, our results suggest that similar domain-specific recipes may be useful in other engineering domains with structured, verifiable outputs. Datasets, models, and inference code are available at https://github.com/AutodeskAILab/floora.

Problem

• architectural layout generation
• domain-specific language model
• foundation models
• structured representation
• human alignment

Innovation

• Domain-Specific Language
• Reinforcement Learning Alignment
• Architectural Layout Generation
• Supervised Fine-Tuning
• Verifiable Rewards