AI Powered Sustainable Façade Design

This experiment develops an AI-powered sustainable façade design engine using HPC. Users input parameters (materials, aesthetics, environmental constraints), and a fine-tuned large language model generates multiple façade configurations. Each alternative is then evaluated by an integrated thermal simulation module scoring heat transfer, solar gain, and energy performance. The work includes building a domain-specific RAG knowledge base, fine-tuning a 70B-parameter LLM via distributed training (PyTorch FSDP, DeepSpeed) on EuroHPC GPU clusters, integrating OpenFOAM/EnergyPlus thermal simulations, and developing a decision-support interface. Technologies used: generative AI, HPC-accelerated LLM fine-tuning, FAISS vector search, and physics-based thermal simulation. The goal is to reduce façade design iteration time from weeks to hours, improve thermal efficiency by 25-30%, and deliver a validated SaaS tool supporting EU Near Zero Energy Building compliance.

Organisations involved: 

End-User: Façade Design Manager

Technology Expert: Conium AI

HPC Provider: TÜBİTAK Bilgem