Design++ Summer School: AI in AEC

Programme

The summer school will consist of lectures, tutorials and practical projects, and will focus on three main topics: generative models for forward and inverse design, machine perception and large language models. The lectures will introduce the foundations for these topics, as well as discuss their applications for architecture, engineering, design and construction. The mini-projects and workshops will provide a practical hands-on experience on selected case studies. The participants will work in small, mix-domain groups on a selected challenge or use-case under the tutors’ guidance.

Forward & Inverse Design: We will focus on applications of supervised machine learning and generative AI models like feed forward neural networks and autoencoders to both forward and inverse design problems. Exemplary application scenarios are surrogate models, sensitivity analysis and generative design, in the field of architecture, civil engineering and similar. Through selected projects, participants will implement solutions based on standard Python libraries such as ScikitLearn or PyTorch, and/or using the open-source tools AIXD and external page ARA, which includes methods for dataset generation, data exploration and model training.

Machine Perception: In the rapidly evolving field of construction, the integration of advanced technologies is crucial for improving efficiency, accuracy, and safety. Join us for an in-depth exploration of 3D world reconstruction in this focus topic. Gain a behind-the-scenes look at the techniques used to create digital world representations, whether for realistic renderings or for the inspection and measurement of geometry-accurate models. Participants will have the opportunity to engage with state-of-the-art approaches in point cloud processing and various computer vision techniques to explore and leverage visual data, creating their own digital twins of the environment or advancing other downstream tasks. Participants will also have the opportunity to experience the whole process, starting from recording their own data, process and interpret these with open-source or commercially available software and state-of-the art approaches.

Large Language Models: This focus topic explores the integration of Large Language Models (LLMs) with domain-specific tools, databases, datasets, and documents. Participants will learn to use LLMs locally through Python, create Retrieval-Augmented Generation (RAG) workflows, and develop simple LLM agents equipped with specialised tools. Participants will work in multidisciplinary groups on projects with specific goals, such as linking a CAD or modelling software of choice with an LLM, tapping into databases for construction materials and products, using PDFs of building regulations for compliance checks, and interpreting computational analysis results.