Adam Gaier
Research scientist
Autodesk Research

"TileGPT: Tile-based Design Exploration with Large Language Models"

Adam Gaier is a research scientist at the Autodesk AI Lab where he pursues research at the intersection of evolutionary and machine learning. He received master’s degrees in evolutionary computing and robotics, and a PhD focused on tackling expensive design problems through the fusion of machine learning, quality diversity, and neuro-evolution approaches. His work has received recognition at top venues across these fields, including a spotlight talk at NeurIPS (machine learning), multiple best paper awards at GECCO (evolutionary computation) and AIAA (aerodynamics design optimization). His current research focuses on the use of large language models and evolutionary optimization for architectural design.


TileGPT, a research prototype developed at Autodesk Research, addresses tile-based design problems through a novel amalgamation of Generative Design (GD) techniques and Large Language Models (LLMs). Utilizing the Wave Function Collapse (WFC) algorithm alongside the diversity-centric MAP-Elites algorithm, the system is capable of producing designs that meet stringent architectural and engineering constraints. These algorithmically generated designs serve a dual purpose: they not only provide immediate utility but also act as a synthetic labeled dataset to train large language models, bridging the gap between machine learning and design in a mutually beneficial loop.

The interface of this system presents an innovative blend of graphical and natural language inputs, enabling users to interact at a high level. They can specify geometric and performance features, generate comprehensive designs, and even modify existing configurations through natural language. This advances the state of the art by integrating Example-based Procedural Design, Diversity-based Optimization, and Language Models into a singular, powerful tool for generative design.

The output from the trained language model feeds back into the WFC system, offering a unique iterative loop where high-level directives from the language model can be refined and constrained by WFC. This is where optimization-based GD plays a pivotal role. Through the MAP-Elites algorithm, which explicitly aims for both diversity and high performance, the system generates a gamut of varied but viable solutions, optimizing starting conditions and initial tiles for WFC.

The uniqueness of this approach lies in its hybrid model that seamlessly interweaves GD and Generative AI (G-AI). GD synthesizes high-quality training data for the language model, and in turn, the language model provides an intuitive layer for on-the-fly design generation. This closed-loop system not only makes generative design more accessible but also opens new avenues for design exploration and validation.

Fundamentally, we are not merely churning out designs but constructing models that comprehend the intricate relationships between geometry and performance. This eliminates the often aimless post-hoc analysis commonly associated with generative design, steering the design process towards a more purposeful and insightful exploration.

In sum, TileGPT contributes to the AI theme of the conference by showcasing how large language models can be effectively integrated into the design process, not as a replacement but as an enhancement. The innovative edge lies in its unique methodology, uniting generative design with language models while solving the data problem, and in its ability to facilitate a novel form of human-AI interaction in the design space


Rob Greig
Global Chief Information Officer

"A Quantum future with AI"


Rob is the leader of Arup’s strategic digital technology ambitions globally. He joined the firm in 2017 bringing over 20 years of experience of technology and digital leadership, digital content development, cloud capabilities and strategic cyber security practice. A seasoned practitioner of leading people through digital transformation, as CIO he heads up a global team based in 90 offices around the globe with 20,000 users. Much of Rob’s work focuses on enabling the firm’s digital transformation, including leading the development of our cloud strategy. He has also led the establishment of the firm’s cyber advisory services.

Prior to joining the firm Rob worked across the creative and public sectors. As Chief Technology Officer of the Royal Opera House he created World Ballet Day, an annual live stream event connecting some of the world’s biggest dance companies with millions of viewers around the globe. As the Director of Parliamentary Digital Services he enabled greater access to information about the UK’s democracy by driving open data initiatives. His work connected the public to political debate by developing the largest social media following of any parliament globally. He introduced pivotal cyber security measures that successfully defended the House of Commons and House of Lords during the 2017 cyber attack.

Rob is a lay trustee for the Royal College of Surgeons, a Fellow of the Institute of Information Technology and a Cultural Fellow of King’s College London. He is formerly a board member of the National Association for Gallery Education (Engage) and a director of the non-profit arts-technology company Tessitura.


Clayton Miller
Associate Professor
National University of Singapore

"How AI Disrupts: Lessons for the Built Environment from Other Industries"

Dr. Clayton Miller is an Associate Professor at the National University of Singapore (NUS) in the Department of the Built Environment.  He is the creator of the edX online Course - Data Science for Construction, Architecture, and Engineering - that has had 30,000+ participants worldwide since April 2020. Dr. Miller's research focuses on performance data analytics using thousands of real-world case study buildings collected from facilities worldwide. 

We all know the hype — but what AI-driven mechanisms do companies actually use when changing how they do business? Numerous examples of AI-driven disruption exist in retail, advertising, travel, manufacturing, and other industries. The built environment sector is slow to innovate, but this gives us a huge advantage to learn from others! This talk will explore the framework of AI disruption in the built environment context. It will cover what AI is good for (point versus application versus system solutions) and explain why humans are still crucial, as they are much better at judgment than AI despite being beaten in terms of prediction capabilities. Applying new data sources from building occupants (wearables, smart devices, etc.) and applications (influencing behavior in buildings) will be covered and explored


Karoliina Torttila
Director of AI
Trimble Inc.

"AI in the Era of Retrofit"

Karoliina leads Trimble AI’s research and development, weaving Deep Learning into a vast technology portfolio spanning from design to reality capture. Her work is dedicated to merging the digital and physical worlds, to reimagine how we design, build and sustain our built environment.


In industrialized nations, most of our built environment is already in place. Approximately 80% of the UK's projected buildings for 2050 are standing today. Japan is actively shifting its focus from building new structures to optimizing and maintaining existing ones. The monumental task ahead? Merging the physical and digital worlds to reimagine how we design, construct and sustain assets in the context of retrofit. Reality capture technology, ranging from smartphones to terrestrial scanners on robots, plays a key role in generating Digital Twin data. However, it's the groundbreaking advancements in AI that unlock the potential of this information.

This talk unpacks AI’s transformative role in:

  • Scene Understanding: Through the scalable interpretation of captured data, AI enables the efficient generation of design starting points and lays the foundation for offsite manufacturing processes.
  • Object Inventory: While identifying objects using Machine Learning is not new, novel algorithms now take it a step further. They can visually pair detected objects with their matches from digital content warehouses. This process results in detailed asset inventories enriched with metadata, positioning them as valuable tools to further the circular economy objectives of reusing and recycling resources.
  • Condition Analysis: Building upon the first two pillars, AI's prowess extends to asset evaluations to inform data-driven maintenance strategies.

Harnessing AI in these areas offers a revitalized vision for the era of retrofit our sector is entering.


Tea ┼Żakula
Associate Professor and the Head of Laboratory for Energy Efficiency
University of Zagreb

"Embracing the Future: AI Integration for Advanced Building Control"

The unstoppable march of technological progress has permeated even the most conservative industries, sparking disruption and innovation. With AI's rapid advancement, the building and energy sectors are now entering a new realm of possibilities. AI's role in the building sector extends beyond mere automation; it holds the potential to revolutionize the way we design, operate, and experience our built environments.  

In this presentation, we share the latest results of our research on advanced building control, employing cutting-edge strategies that leverage AI technology. Here, AI plays a pivotal role in executing model predictive control (MPC) and personalized comfort models, bridging the gap between technology and humanity. It's a delicate dance where humans actively participate, offering unique preferences, constraints, and invaluable feedback, while AI's optimization capabilities tackle the multifaceted challenges posed by competing objectives.  

However, the true challenge lies in implementing these theoretical advancements in the real world, where tangible demonstration sites remain extremely scarce. In this talk, we present our latest findings from the implementation and testing of AI-based algorithms at the RCK Rudera Boskovica living laboratory for intelligent buildings in Zagreb, Croatia. Just as AI generated an abstract for this talk using limited inputs, our goal is to make use of limited data acquired on a building to optimize building operation with as few inputs as possible. 

What do people truly desire from AI? How do they wish to interact with this new force? And, most importantly, to what extent can AI genuinely assist us in operating our buildings, and what might be the price we pay for such invaluable assistance? These thought-provoking questions form the focal point of this talk as we navigate the uncharted territory of AI's potential impact on our built environment, unraveling the complex relationship between humans and AI. Ultimately, this presentation serves as a platform for stakeholders in the AI, building and energy industries to explore the practical integration of AI in buildings, engaging in a dialogue that shapes the future of our built environment.



PR partner

Organized by

Supported by