MuNa/ref/buildings-15-02400-v2.txt

Article

LLM and Pattern Language Synthesis: A Hybrid Tool for
Human-Centered Architectural Design
Bruno Postle 1

and Nikos A. Salingaros 2,3, *
1
2
3

*

Union Street Research, 18-20 Union Street, Sheffield S12 JP, UK; bruno@postle.net
Department of Mathematics, The University of Texas, San Antonio, TX 78249, USA
Thrust of Urban Governance and Design, Hong Kong University of Science and Technology (Guangzhou),
Guangzhou 511453, China
Correspondence: salingar@gmail.com

Abstract
This paper combines Christopher Alexander’s pattern language with generative AI into
a hybrid design framework. The result is a narrative synthesis that can be useful for
informed project design. Advanced large language models (LLMs) enable the real-time
synthesis of design patterns, making complex architectural choices accessible and comprehensible to stakeholders without specialized architectural knowledge. A lightweight,
web-based tool lets project teams rapidly assemble context-specific subsets of Alexander’s
253 patterns, reducing a traditionally unwieldy 1166-page corpus to a concise, shareable
list. Demonstrated through a case study of a university department building, this method
results in environments that are psychologically welcoming, fostering health, productivity,
and emotional well-being. LLMs translate these curated patterns into vivid experiential
narratives—complete with neuroscientifically informed ornamentation. LLMs produce
representative images from the verbal narrative, revealing a surprisingly traditional design
that was never input as a prompt. Two separate LLMs (for cross-checking) then predict the
pattern-generated design to catalyze improved productivity as compared to a standard campus building. By bridging abstract design principles and concrete human experience, this
approach democratizes architectural planning grounded on Alexander’s human-centered,
participatory ethos.
Academic Editor: Li Yang
Received: 18 June 2025

Keywords: adaptive design; AI-driven architecture; Christopher Alexander; evidencebased design; human-centered architecture; large language model; pattern language

Revised: 1 July 2025
Accepted: 6 July 2025
Published: 9 July 2025
Citation: Postle, B.; Salingaros, N.A.
LLM and Pattern Language Synthesis:
A Hybrid Tool for Human-Centered
Architectural Design. Buildings 2025,
15, 2400. https://doi.org/10.3390/
buildings15142400
Copyright: © 2025 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license
(https://creativecommons.org/
licenses/by/4.0/).

Buildings 2025, 15, 2400

1. Introduction
The dominant architectural culture throughout the 20th and 21st centuries that embraced industrial minimalism dismissed ornament as redundant or superficial. However,
recent findings in biophilic design, neuroscience, and psychological research reveal ornamentation’s essential role in human health and cognitive functioning [1–3]. Implementing
Christopher Alexander’s human-centered A Pattern Language with adaptive ornamental
design presents a promising alternative. By leveraging large language models (LLMs),
complex decision-making is synthesized while enhancing user health and well-being
through design.
Alexander et al.’s A Pattern Language (1977) [4] represents one of the most comprehensive attempts to codify design wisdom for the built environment, offering 253 patterns
that span from regional planning down to interior details. The complete pattern language

https://doi.org/10.3390/buildings15142400

Buildings 2025, 15, 2400

2 of 32

spans 1166 pages, making it unwieldy for a “quick” practical application. Additionally, the
traditional method of applying these patterns typically requires specialized architectural
knowledge, creating a barrier between experts and the stakeholders who will ultimately
inhabit these spaces.
In response to these challenges, this article combines digital tools with artificial intelligence to make Alexander’s pattern language more accessible and applicable. Traditional
applications of the pattern language or purely computational approaches applied separately
fall short in managing complexity or enhancing emotional understanding. The present
method consists of two key components:
1.
2.

A web-based application enables the creation of project-specific pattern subsets
through an interactive hypertext interface.
The use of large language models (LLMs) synthesizes these pattern subsets
into narrative descriptions that communicate the experiential qualities of the
proposed architecture.

This technique aims to bridge the gap between abstract architectural patterns and
concrete spatial experiences. Stakeholders can better understand and evaluate proposed
architectural solutions before construction begins through narrative descriptions that translate design patterns into vivid, experiential language. Pattern language coupled with
generative AI transforms the relationship between architectural professionals and the communities they serve. The method shifts architecture from the pursuit of visually striking
forms to the systematic creation of neurologically and psychologically healthy environments. The primary metric for architectural quality is empirically measured cognitive
performance and well-being.
There are broader issues that detach the design profession from common everyday life.
An expertise barrier nowadays results in stakeholders having a limited understanding of
and input into the architectural decisions that will profoundly affect their daily experiences.
The planning process becomes opaque to laypeople, remaining the exclusive domain of
design professionals. Consequently, stakeholders may find themselves unable to meaningfully influence design decisions that directly impact their lives [5]. The hybrid AI—pattern
language method makes complex architectural decisions accessible and understandable to
non-specialists.
Choosing a building meant to house a university department of Computer Science
and AI as one example illustrates how the method works in practice (Section 4). The
results differ substantially from the standard top-down method of designing such a project.
These differences underline the very desirable adaptive qualities of the present model.
Human-centered design turns out to improve the users’ health and well-being in the long
term [6]. This assessment is verified here by using two separate large language models to
predict the comparative productivity of the department, based entirely on the building’s
architecture (Section 5).
In this second part of the paper, ChatGPT (o4-mini-high) evaluated the university
building resulting from the pattern language versus standard new campus buildings, and
the LLM made an astonishing claim: “the fashionable buildings’ stress-inducing geometries
would predict lower publication rates, fewer patent applications, and a shift toward safe,
incremental research rather than bold, exploratory ventures.” To cross-check this evaluation,
a separate prompt to ChatGPT-4o used a different account and obtained a very similar
response (Section 5).
Generative AI’s conjecture on improved academic productivity based solely on architectural design is unexpected. This claim challenges prevailing views of how physical spaces influence human cognitive and emotional responses. The hybrid AI–pattern
language approach redefines architectural quality itself by enhancing human cognitive–

Buildings 2025, 15, 2400

3 of 32

emotional health and productivity—not as abstract forms provoking esthetic novelty.
Adaptive design lies not in following rules rigidly or ignoring them entirely but in working
creatively with patterns as constraints.
Generative AI through an LLM predicts that the everyday psychophysiological experience of an academic building will result in measurable differences in productive research
from the department. Are we witnessing a “Move 37” moment in architecture? (Referring
to the historic Go match between AlphaGo and Lee Sedol in March 2016) [7]. If that is the
case, then the profession should pay attention. Using focused and intelligent prompts, the
LLM can give a more unbiased assessment than humans normally can because the prompts
can steer it around cultural, intellectual, and media pressures (Section 7).
Design patterns are not taught in architecture schools, so students are unaware of
them. Applying the pattern language requires training to master most of the patterns so
that they can be combined and recombined in reaching an adaptive, bottom-up design
through iteration [8–10]. Today’s schools value the spontaneous, top-down creation of
visually appealing designs judged by their novelty. Cognitively, this privileges the simplest
possible path to design that avoids the extensive recombination of complex components.
After several years of training in this direct method of generating designs, young architects
tend to lose the mental ability to organize complexity by adopting what is known as “design
fixation” [11].
By challenging architectural culture’s application of “design-through-image”, this
paper avoids the limitations of mainstream approaches that prioritize visual abstractions
over cognitive and emotional human experience. Large language models (LLMs) translate
tested design patterns into emotionally engaging narratives that bridge theory and practice
in a novel way. In a significant shift, the model reintroduces neuroscientifically backed
ornamentation as being integral—not merely decorative—for enhancing physiological and
psychological well-being.

2. Literature Review and Background Problems
2.1. Alexander’s Pattern Language and Its Impact
A Pattern Language emerged as a revolutionary approach to architectural and urban
design. The work presented 253 patterns arranged hierarchically from the largest scale
(regions and towns) to the smallest (building details and ornament). Each pattern describes
a recurring problem in the built environment and offers a solution that can be adapted
to specific contexts. Significantly, Alexander conceived of these patterns not as isolated
elements but as an interconnected language, with each pattern linking to higher and lower
patterns in the hierarchy.
Christopher Alexander (1936–2022) trained in both Architecture and Mathematics
at Cambridge and obtained the first PhD in Architecture ever awarded by Harvard. He
was one of the first people to apply computers to architecture, in the 1960s, trying to
manage complexity in design theory. His writings combine computation with embodied
understanding, and for this reason, they have influenced both architecture and computer
science. Alexander practiced architecture, designing many buildings around the world.
(The second author, N.A.S., is the principal editor of Alexander’s four-volume book, The
Nature of Order).
A Pattern Language introduced a holistic framework for architectural design, advocating interconnected solutions adaptable to context and scale. Its practical application
was limited by the profession rejecting some of its key features, but that is now changing.
Contemporary research in biophilic design highlights ornament’s role in psychological and
physiological health [12–14]. Techniques such as eye-tracking and visual attention software
empirically demonstrate how ornamented environments support cognitive and emotional

Buildings 2025, 15, 2400

4 of 32

well-being, in sharp contrast to minimalist aesthetics, which often induce psychological
disengagement and stress [15–18].
Design patterns apply to all possible structures: the human use of buildings and urban
spaces is influenced by an embodied understanding that connects the user to the physical
setting. Therefore, the present method adapts to designing different types of buildings
(such as commercial, medical, and residential), thereby enhancing the general value of the
research. Each type of building will have different functional requirements and specifics,
and the method is entirely general and able to handle an infinite variety of projects. The
present description is detailed for a specific example, but only to illustrate how the method
works in practice.
The impact of Alexander’s work extends far beyond architecture. The pattern language approach has influenced fields ranging from software design [19] to education [20]
and organizational development [21]. The enduring influence of his work speaks to its
fundamental insight: that complex design problems can be addressed through a combinatorial and modular language of solutions that connect across domains and scales. A
selection mechanism evolves design combinations by adapting them to human emotional
and physiological needs, not abstract images.
2.2. How Design Patterns Circumvent Design Through Images
Patterns are compressed verbal descriptions of recurring visual—actually, sociogeometric—relationships [4,6,8–10]. For centuries, architects have relied upon images
to communicate architectural ideas and solutions, and this has become a cognitive working
standard. Colleagues who judge architecture primarily through images look for formal
precedents and visual proof. Since the present approach is totally narrative-based, readers
might be confused about the proposed methodological framework.
Each design pattern discovered by Alexander and his colleagues in the 1970s was
initially tested heuristically for its adaptivity to human emotional well-being. This was the
principal criterion for choosing from among a much larger set of design pattern candidates
which ones to include in the book A Pattern Language. Since its publication, the success
of the patterns has been repeatedly verified by the feedback that pattern-based buildings
provide to their users.
The present article is based upon narrative and process, not images. The operational
sequence is as follows: design pattern repository → specific pattern subset → LLM verbal
narrative → human-centered design criteria → evaluation; this describes the results entirely
in words. For this reason, readers familiar only with image-driven design will not see the
familiar visual chain of drawings → renderings → photographs that, in their mental model,
represents the standard design method.
2.3. Challenges in Pattern Language Application
Alexander’s A Pattern Language requires special methods to be effective in practice.
Self-builders find a useful resource in picking a handful of design patterns to apply to their
project, but the comprehensive nature of the work makes it unwieldy for “easy” use in
a more substantial task [22]. A project needs to synthesize and combine several design
patterns, so implementation becomes an exercise in organizing complexity, which is a
non-trivial problem. Furthermore, the interpretation and application of patterns typically
require a minimum of familiarity with the patterns.
The traditional method of applying a pattern language involves reading and understanding the entire work, identifying relevant patterns through expert judgment, and
keeping these patterns in mind during the design process. Several iterative steps at combining the patterns to generate adaptive forms require doing this in one’s head, with the

Buildings 2025, 15, 2400

5 of 32

limited help of visual aids. The double process of pattern selection and combination remains challenging to non-experts (including most architects trained in the design-throughimages paradigm).
2.4. Digital Tools and Pattern Language
Various attempts have been made to digitize and make the textual description of
pattern languages more accessible. As well as producing 80 additional patterns in A
New Pattern Language (2020) [23], architectural theorist and urbanist Michael Mehaffy has
advocated for digital adaptations of pattern language to enhance its usability. Mehaffy
worked extensively with Christopher Alexander. Projects such as Iba’s Pattern Language
3.0 [24] and Schuler’s Liberating Voices pattern language project [25] have explored digital
formats for pattern language solutions in various domains.
However, these efforts have typically focused on creating comprehensive digital repositories rather than tools for implementing context-specific pattern languages. The approach
presented in this article differs by emphasizing the creation of manageable, project-specific
pattern subsets rather than attempting to manage the entire pattern language. An additional and non-trivial obstacle is that Alexander’s original A Pattern Language is not open
access, thus it cannot be posted freely on the web.
While the authors have been very careful to respect other authors’ rights in this paper,
the proposed method could be implemented entirely without using the canonical Alexandrian patterns (Section 9.2). LLMs can approximately duplicate the design patterns strictly
from secondary sources. The errors are too high to make that a useful method, however,
since the model will never be able to regurgitate an accurate facsimile of Alexander’s
pattern language.
2.5. Language Models in Architectural Contexts
Large language models (LLMs) offer new possibilities for architectural communication
and planning [26,27]. Some architects are exploring the potential of LLMs to generate
architectural descriptions, investigating conceptual design phases towards visual implementations of novel forms rather than adaptation to human affordances and scale [28]. The
second author (N.A.S.) has applied LLMs to describe environments for creative work [29]
and to classify window typologies that generate anxiety [30].
The approach adopted here specifically focuses on how LLMs can translate pattern
languages into experiential narratives that communicate architectural qualities to nonexperts. Alexander’s original motivation for the pattern language was indeed to bring the
design process closer to common people, and this is the reason for its continued success
with self-builders. The present application addresses what Tzonis identifies as a persistent
challenge in architectural communication: the gap between abstract design principles and
the lived experience of architecture [31].
Anticipating a possible misunderstanding, the input of design patterns into the present
method is through their descriptive text, not from images. The authors do not use example
Figures and images to explain the pattern language to the AI. While the book A Pattern
Language includes one image (photo) to illustrate the “feeling” of each pattern, these are not
included in the software used by the present model; only the re-worded verbal description
of the solution or its paraphrase is included. The pdf short-list prompt therefore comes
from the pattern statements only.
In the authors’ opinion, architecture has accumulated cognitive scaffolding over time
that constrains adaptive development. Design that always works within an image-based
method may never develop autonomous capabilities, thus precluding adaptive problemsolving. Human-centered innovation involves deliberately sidestepping these supports. It

Buildings 2025, 15, 2400

6 of 32

therefore makes sense to develop AI as a tool that enhances collaborative intelligence, so it
exceeds the capabilities of individual human efforts. Developing technologies that enable
true cognitive symbiosis will have to work with human cognitive mechanisms.
2.6. Stakeholder Participation in Architectural Design
Numerous scholars have emphasized the importance of stakeholder participation in
architectural design, including Till [32], who argues for a more democratic approach to
architectural practice. Sanoff’s fundamental work on participatory design [33] highlights
the value of involving end-users in the design process, developed further by Salama [34],
while Blundell Jones et al. [35] document various approaches to architecture as a social
practice. While much useful discussion on participatory design has taken place over the
years, no satisfactory method has emerged that the building industry has felt comfortable in
adopting. For this reason, design and construction have continued to implement standard
typologies without user input.
In a welcome development, AI offers new solutions to participatory design [36,37].
This is due to AI’s ability to handle the complexity of multiple decision-making processes and to combine them into a suitable result. The integration of AI-generated narratives fundamentally democratizes architectural design by breaking radically from imagedriven methodologies.
Using AI potentially enhances stakeholder understanding and participation in the
design process. Design patterns represent evolved architectural and urban solutions,
invented by ordinary builders and selected by the general population. They arise from
common practice in each society, thus forming an essential part of material culture [38]. By
making pattern language more accessible and translating pattern groupings into concrete
narratives, this paper addresses what Friedmann [39] identified as the “knowledge gap”
that often limits meaningful participation in urban planning processes. Curiously, therefore,
AI makes possible human interaction and participation in the design process that was
unwieldy or impossible before.
2.7. Exploiting Feedback Loops to Improve AI-Based Results
Here, LLMs generate narratives and then again LLMs evaluate their effectiveness,
which constitutes cognitive “self-circulation”—the model evaluates its own output and
lacks independent third-party evaluators. Attempts are made to use different accounts
and models for cross-validation, yet this fundamentally does not depart from the issue of
model self-referencing, which is insufficient to support causal claims. While admitting to
this valid criticism, the feedback procedure is in fact a key asset in AI-based methods.
AI systems can significantly enhance an iterative design and development process by
leveraging feedback loops to continuously improve results. Recursion involves using AI
to analyze feedback, identify areas for improvement, and then refine the result based on
those insights. A cycle of feedback, testing, and refinement allows for rapid recursion—
impossible or impractical for humans—and ultimately leads to more effective AI solutions.
It is precisely the iterative cycle method that was used to derive protein folding, earning
the 2024 Nobel Prize in Chemistry.

3. Methodology: A Listing of Design Patterns
3.1. Development of the Web-Based Pattern Subset Tool
The model begins with a web-based application designed to make Alexander’s A
Pattern Language more navigable and accessible. A private and non-commercial application
titled “APL-Companion” presents each of the 253 Alexandrian patterns in a collapsible
format. The problem/solution content is new text that has been written entirely for this

Buildings 2025, 15, 2400

7 of 32

application. This application curates project-specific subsets of Alexander’s patterns,
simplifying stakeholder choice and interaction. Readers can find the application from the
information provided in Appendix A. This functionality is well within the capabilities of
LLM coding agents. A brief technical specification for the tool follows:

•
•
•
•
•
•
•
•
•
•

Collapsible interface: Each pattern is contained in HTML <details> elements that can
be opened/closed.
Smart URL encoding: Open patterns are encoded in the URL fragment (e.g., # p = 1,
3–7, 12) using compact range notation.
State persistence: Selected patterns remain open when returning to bookmarked URLs.
Cross-references: Links between patterns automatically highlight when target patterns
are open.
Position memory: When clicking pattern links, the tool remembers scroll positions
and returns users to their previous location when closing patterns.
Auto-scrolling: Automatically scrolls to newly opened patterns for smooth navigation.
Visual feedback: Links to currently open patterns are visually distinguished.
Pure JavaScript: No external frameworks, using modern browser APIs.
Responsive design: Mobile-friendly layout with touch-optimized controls.
Print optimization: CSS print styles hide navigation elements and show only
selected content.

Allowing users to expand and collapse individual patterns creates a more manageable interface for navigating the comprehensive pattern language. The first author (B.P.)
developed the pattern subset tool as a single-page web application using HTML, CSS, and
JavaScript. The application’s core functionality centers on the HTML “details” element,
which provides native browser support for expandable/collapsible content sections. Patterns are implemented as separate “details” elements, allowing users to toggle the visibility
of individual patterns. The application maintains a lightweight footprint, requiring no
server-side processing or database. All functionality is implemented client-side, making it
easily deployable on any static hosting service.
Manual user selection creates a specific subset pattern language, with expanded
patterns constituting the subset. The selection process typically takes minutes rather than
the hours or days that might be required to read and process the entire pattern language
(not to mention that only someone already very familiar with each pattern is capable
of doing this easily). The resulting subset represents the patterns deemed relevant to a
particular building project.
Interested readers are encouraged to follow the general outline set out here to implement a parallel scheme for organizing the patterns for convenience. All that is required is a
selection of patterns (containing only the pattern title, problem, and solution) deemed to be
relevant to a specific project. The pattern list is then fed to an LLM as a PDF file along with
the correct prompt to generate the narrative (this is detailed below).
3.2. URL Fragment Approach for Creating a Subset Pattern Language
To enable the sharing and persistence of selected pattern subsets, the model implements a URL fragment approach. The application stores the selected pattern subset as
URL fragments, allowing users to bookmark or share specific pattern subsets via links.
When accessed, these links automatically retrieve the selected pattern subset, ensuring
consistency across different users and sessions. The user interface presents patterns in
their hierarchical order, from the largest scale (regions and towns) to the smallest scale
(building details). Each pattern is represented by its number and title when collapsed,
with the expanded view showing a concise summary of the pattern along with links to
related patterns.

Buildings 2025, 15, 2400

8 of 32

The method preserves Alexander’s concept of patterns as an interconnected language
through hypertext navigation. Each pattern includes links to higher patterns (which it
helps to implement) and lower patterns (which help implement it). These relationships
form what Alexander described as a “network” of patterns that work together to create
coherent design solutions. The methodology proceeds as follows: (1) A designer who
is familiar with the different patterns in A Pattern Language selects the titles of all the
possible patterns that appear relevant to the project. In this case, the example chosen is a
university building meant to house the Department of Computer Science and AI. (2) The
APL-Companion software generates a detailed textual description for use as a prompt with
LLMs, as described next.
For example, a subset pattern language for a university department might be encoded
as follows, selecting design patterns by their number:
file:///C:/Users/Username/Documents/apl.html#p=18,80,82,88,95–96,98–99,102,
107–108,110,112,115,119–120,122,124–125,127–130,132–133,135,146–148,150–152,159–161,
163–164,166,171,174,176,179–180,183,191–192,194,207,222–223,225,232–233,235–243,248–250
This approach eliminates the need for server-side storage or databases while ensuring
that pattern subsets can be easily shared among stakeholders.
Simply holding all the selected design patterns in one’s mind when working on a
project (as was necessary for previous implementations) is a very challenging cognitive task.
It would normally require weeks of familiarization and working with the list repeatedly
to grasp a global synthesis. The necessary next step of recombining patterns to approach
more optimal results is even harder. For this reason, joining a pattern language to a large
language model is a huge step in being able to manage the combinatorial complexity that
an adaptive design process requires.
3.3. APL-Companion Generates a PDF Pattern List for LLM Context
The APL-Companion application supports printing functionality, typically to PDF, that
includes only the pattern subset titles and summaries without extraneous elements such
as navigation links. This condensed output provides an ideal context for LLM prompting,
offering the relevant pattern information in a format that can be directly input to large
language models. This format ensures that the LLM receives clear, relevant information
about the selected patterns without being overwhelmed by the complete pattern language.
For reference, the list of design patterns by the number and title selected for this
project, a university building to house the Department of Computer Science and AI, is given
in Appendix B. Selecting these 65 patterns manually involves a subjective decision; yet
researchers familiar with the pattern language would likely choose a similar list containing
most of them. Surprisingly, application to an entirely different project involves changing
only a few of these patterns, since around 50 of them help to establish an embodied
understanding for the user, valid for any building of similar size. The prompt for a different
project will of course generate an entirely different design narrative that might require
some new patterns.
The brief descriptions of these design patterns are not included here. The APLCompanion web application paraphrases and substantially rewrites the content of all the
patterns so they are not identical to the official published text. The educational aim of this
application is not to publish the selected pattern list containing a factual summary but to
feed it as a prompt into an LLM. What is published is the ensuing AI output, which creates
a descriptive narrative for the project.

Buildings 2025, 15, 2400

9 of 32

3.4. LLM Integration and Prompt Engineering
The second phase involves using large language models to synthesize narrative descriptions based on the selected pattern subsets. LLMs translate these pattern subsets into
coherent, experiential narratives and adaptive ornamental solutions guided by neuroscientific criteria. For this research, the authors utilized Claude 3.7, an advanced language
model capable of processing substantial context and generating coherent narratives. There
are several equivalent LLMs that could be used in this manner, so this is only a convenient
choice adopted for writing this paper that should not affect the study’s reproducibility
or generalizability.
The basic prompting process involves providing the LLM with the pattern subset
(typically in PDF format) along with specific instructions regarding the building project.
These instructions include the following:
1.
2.
3.
4.
5.

The purpose of the building (e.g., a university department).
The approximate size or capacity of the institution (e.g., 200 students and staff).
Any specific local requirements or contextual factors.
A request for a narrative description focusing on experiential qualities.
An explicit mention including the ornament.

A typical prompt structure follows this general format:
Prompt: “Attached is a pattern language for [specific building type], this is a [size
description]. The [building/institution] is [purpose description]. Write a narrative description showing how the building is experienced, describe the look and feel and the
ornamental treatment.”
3.5. Eventual Need for New Patterns—LLMs Greatly Simplify the Task
Any project designed with the pattern language will normally require additional
patterns to be developed that are not already included among the 253 canonical design
patterns. These supplemental patterns will address design problems specific to the project
and could play an important role [8,9,40]. Some patterns could be selected from among the
80 in Mehaffy et al.’s A New Pattern Language, while others need to be newly discovered.
Fortunately, A New Pattern Language is open-source and is published on two separate sites
in slightly different formats as well as in book form [41]. The 80 new patterns can be quickly
examined, and the URLs of any relevant ones are included in the prompt to the LLM.
The task of writing a few entirely new patterns is necessary to ensure an optimal
design outcome. This is a separate topic that will not be developed here: for simplicity, this paper uses only the original pattern language. Michael Mehaffy is working on
a project using LLMs to derive new patterns [42]. Preliminary results reveal that the
process is enormously facilitated by AI, reducing the considerable amount of work traditionally required to discover a new design pattern. With the facility of assembling data
provided by an LLM, the normally laborious and time-consuming derivation becomes easy
and straightforward.
Looking to the software community for useful lessons reveals fervent activity that
combines LLMs with design patterns. But so far, interest mostly focuses on deriving
and applying design patterns that improve AI functionality, such as in optimizing and
organizing prompts. What is meant here is the reverse: using generative AI to derive new
design patterns. Interest lies in architectural design, yet the concepts in computer science
are similar. The study by Nazar et al. goes in the desired direction [43]. A paradigm shift is
expected when an LLM-based program transforms the project’s nature by simultaneously
applying documented design patterns and discovering entirely new ones.

Buildings 2025, 15, 2400

10 of 32

4. Case Study: A University Department Building
4.1. General Features Emerging from the Use of the Pattern Language
A university building housing a Computer Science and AI department demonstrates
this adaptive design method. Stakeholders will collaboratively select patterns reflecting needs such as communal spaces, human-scaled environments, and natural lighting.
Using LLM-generated narratives and neuroscientific insights, ornamental elements are
strategically designed to support psychological well-being [6,29,30]:

•
•
•

Exterior façades feature ornamented entrances and structural/visual frames employing fractal scaling to induce positive subconscious engagement.
Interior spaces incorporate ornamental panels with plant-like, fractal designs, enhancing cognitive function, particularly in learning environments.
A monumental staircase is designed with ornamental complexity, emphasizing natural
lighting and visual stimuli conducive to memory and emotional well-being.

These design requirements were not input separately but arose as an essential part
of the pattern combinations. Design patterns already contain them because the pattern
method embodies an embodied, human-centered approach to design that reuses discovered
solutions. Because of this essential adaptation, pattern-generated designs tend to look more
“old-fashioned” than contemporary or industrial designs—but not because they copy any
traditional images.
The web application created an appropriate pattern subset and generated a narrative
description using the LLM. This theoretical application provides a context for testing the
technique’s flexibility and efficacy. To ensure fidelity to Alexander’s patterns and identify
any LLM deviations or hallucinations, any elements in the narrative that are not supported
by the selected patterns were flagged. When discrepancies were found, the narrative
was regenerated until it aligned with the pattern subset. This iterative approach ensured
that the final narrative authentically reflects the selected patterns rather than introducing
extraneous concepts from the LLM’s training data.
4.2. Descriptive Narrative Generated by the Large Language Model
In what follows, the extracts of a detailed description of the human experience of a
user in the proposed building are included. Generative AI makes this conjectural narrative
possible. The text below is selected from a much lengthier description provided by the
large language model, Claude 3.7. The unedited narrative is included as Appendix C at the
end of this paper, since it will be used for the LLM analysis in the second part.
Prompt: “Attached is a pattern language for a new university department of computing and AI, this is a small institution with no more than 200 students and members of staff
at any one time. The department is a venue for human flourishing and creating humane
technologies via a building that meets human needs. Write a narrative description showing
how the building is experienced, describe the look and feel and the ornamental treatment.”
Response (condensed and edited): Narrative description of a Department of Computing and AI—Human-Centered Design.
“The new Department of Computing and AI rejects the usual monumental “campus box”. Instead of a single slab, a cluster of small, inter-linked wings nestles into the
landscape; four storeys or less, so the scale remains human. A carved door and arched
reveal communicate a welcome signal rather than institutional impersonality. A timberframed main entrance modestly projects from the façade, while smaller doors nearby form
a “family”, so visitors can intuitively read the building without relying on signage.
Upon entering the building, the space opens into a sun-filled atrium. Daylight entering
from many directions is filtered by hanging plants; staircases embrace the edges, doubling
as casual seating so circulation and social life merge. Students and faculty are always

Buildings 2025, 15, 2400

11 of 32

in visual contact with the wider community. Teaching zones are kept under 25 ft wide,
so every desk enjoys daylight and views. The largest studio uses a staircase-as-stage,
allowing teaching or impromptu encounters. Alcoves, deep window seats, and narrow
side rooms give introverts privacy within arm’s reach of collaboration—impossible in the
rigid corridors of conventional blocks.
Research workshops occupy the upper levels and side wings, where small teams can
tune their own environments with movable partitions and pin-up walls. Many 3–8-person
meeting rooms encourage serendipitous conversations. A student can shift from coding
sprint to contemplative reading. Three distinct courtyards—one sunny and social, one
trellised and contemplative, one pergola-covered “outdoor classroom”—lie no more than a
minute’s walk from any desk. Low sitting walls and column-marked corners create outdoor
“rooms” rather than leftover voids. Small-paned windows in deep, splayed reveals help to
moderate glare and invite occupation of the windowsill.
Soft plaster walls, wood or cork floors, and half-inch trim lines wherever materials
meet guide the hand and sustain the building’s warmth. Pastel and rich colors and
geometric ornament—etched glass, carved lintels, patterned floor tiles—add detail that
rewards close attention. This contrasts sharply with the blank metallic surfaces of a
typical tech faculty. The building gives occupants a rich spectrum of spatial choices,
continuous connection to nature, and a legible, craft-infused environment. Where the
standard university block impresses but isolates, this pattern-generated design connects
and supports human spaces.”
Since architects tend to be visually oriented, the verbal narrative was converted into
representative images, just to illustrate what is going on. These images are “proof-ofprinciple” only and should not influence details of the eventual design. Using the full narrative from Appendix C (not the shortened version presented above), ChatGPT o3 was asked
to generate some exterior and interior views. No particular style was imposed—certainly
not a Classical or traditional one. ChatGPT was supposed to use only the descriptive
narrative and nothing else. Those resulting images are shown as Figures 1–5.

Figure 1. Main courtyard of the campus building. Image generated by ChatGPT o3 prompted by
N. Salingaros.

Buildings 2025, 15, 2400

12 of 32

Figure 2. Secondary courtyard of the campus building. Image generated by ChatGPT o3 prompted by
N. Salingaros.

Figure 3. Main interior space of campus building. Image generated by ChatGPT o3 prompted by
N. Salingaros.

Figure 4. Hall with ornamented staircase and floor. Image generated by ChatGPT o3 prompted by
N. Salingaros.

Buildings 2025, 15, 2400

13 of 32

Figure 5. Window seats. Image generated by ChatGPT o3 prompted by N. Salingaros.

The description may appeal to a university administration looking for a new building
to house its Computer Science and AI Department, yet the interesting result here is the
humanity of the narrative. Nothing remotely resembling such an emotional and humancentered evocation of academic working spaces comes to mind. And, surprisingly, it took
AI to generate it. By contrast, the standard architect-generated narrative for an equivalent
project seems concerned mainly with formalism and visual effect. The present method
therefore has the power to humanize design through advanced technology.
The use of a pattern language turns design into the realm of the evocative and the
sensory, and away from the industrial and mechanical. This is going against forces that
have pushed architecture in that direction for one century. A different philosophical and
methodological approach generates a narrative from human feelings coming from patternbased forms and spaces. It is time to consider what an embodied understanding of the
built environment offers to design.
Upon seeing Figures 1–5, architects assumed that the present method prompted AI
to generate a Classical/traditional building. They missed the logical sequence of design
steps. This incorrect reaction is due to their intensive training in the design-through-images
paradigm. To reiterate, the model follows the process: design pattern repository (verbal)
→ specific pattern subset (verbal) → LLM verbal narrative. The images were generated
by ChatGPT from the LLM’s verbal narrative. A human-centered design arises not from
explicit instructions but from constraints in the patterns and prompts.
The existing “knowledge” of an LLM is not entirely trustworthy, but what is wonderful
is the ability to feed it “context”. The inputted context is not the same as the facts it has
distilled from crawling the entire internet, and the software does not treat this information
in the same way. The power is that one can supply information and ask the LLM to use its
full and complete understanding of the way language works to transform it into something
that is valuable. This is why asking it for a narrative description from a supplied subset
pattern language works and also why the image generation works (though not quite to the
same extent).
4.3. Multimodal Empirical Validation
An additional step comprises an essential part of the complete design method
but will not be carried out here. Eventually, one needs to test whether the narrative
spaces really promote user well-being and creativity (Figure 6). This can be accomplished with biometric tools and eye-tracking metrics during the desktop viewing of
VR mockups to measure the reactions of human subjects to visuals created from a verbal

Buildings 2025, 15, 2400

14 of 32

description [15,17,29]. The verbal narrative can also be tested using specific response variables (heart-rate variability, etc.) so that the causal chain “geometry → affect → creativity”
becomes testable. Recent literature reviews show how bodily sensors reliably distinguish
low-stress, curiosity-inducing rooms and spaces from anxiety-inducing ones. Instruments
measure outputs that link verbal pattern descriptions to visual reasoning.

Figure 6. Diagram of the hybrid, text-based workflow linking design patterns, LLM engine, and
empirical evaluation. A recursive loop successively improves the design narrative through evidencedriven adjustment of pattern selection. Drawing by N. Salingaros.

A separate verification method is to generate realistic visuals from the descriptive
narrative, such as Figures 1–5, and use a VR walk-through for several participants. Such
experiments are not undertaken here, as they exceed the scope of this paper. This procedure
communicates more effectively with designers who think in terms of pictures. Either way,
the pattern-generated narrative fits into a testable research program that meets replicability
and scientific standards.

5. Conjectured Creative Output from Comparable Academic Buildings
5.1. Does a Building’s Architecture Determine the Creative Work Inside?
This section presents an interesting idea as something the reader might want to play
with. The LLM evaluates pattern subsets through neuroscientific criteria, ensuring fidelity
to Alexander’s principles and optimizing emotional engagement through ornamentation.
AI is therefore applied in two separate stages: (1) to assemble design patterns, then extract
a narrative from them; (2) to diagnose the value of the result by comparing it with standard
architectural typologies. Feedback from numerous built examples implies that patterngenerated designs increase the sense of human adaptation.

Buildings 2025, 15, 2400

15 of 32

In the rest of this paper, a large language model compares the pattern-based building
to fashionable academic buildings meant for the same purpose. The second author (N.A.S.)
uploaded the LLM-generated narrative given in Appendix C along with the prompt to
evaluate how “good” the building was compared to others. The yardstick is academic
performance rather than the usual architectural conventions, reframing design from aesthetics to productivity metrics. To assure objectivity, this investigation used a different large
language model, ChatGPT o4-mini-high, accessed from a different account and computer
in a new chat. Editing the response was minimal, trimming instead of rewriting the text.
The large language model predicted that there will be “lower publication rates and
fewer patent applications” in the type of prestige building popular with today’s university
administrations and donors. The response casts into doubt how contemporary architecture
envisions and implements design thinking for creative work environments, such as those
found in university departments. Readers can draw their own conclusions from these
disturbing answers. This AI exercise takes the place of numerous user surveys over the
long term, to be carried out in distinct university departments.
Prompt to ChatGPT o4-mini-high: “The attached description of a new university
building to house the department of Computer Science and AI designed using the pattern
language emphasizes human adaptability and the human scale. Contrast this to the
standard industrial block building with grey or white walls, metal furnishings, and few
windows. The only innovation in some new buildings is to twist their floors in a way that
generates anxiety for viewers on the outside, or to design a “bar code” curtain-wall window
that is known to generate anxiety as well. Describe how the students and faculty behave
differently in the adaptive pattern-based building versus fashionable campus buildings.
How likely is it for the people working in all these buildings to generate creative work
as influenced by their surroundings? Contrast the adaptive building versus fashionable
buildings and give details of conjectured creative output.”
Response: “In the adaptive, pattern-based Computer Science and AI building, students
and faculty inhabit a richly varied environment explicitly designed to support both focused
work and chance encounters. Spaces like these align with causal frameworks showing
that environments which (i) facilitate social “encounters” and serendipitous interactions
and (ii) offer abundant “views” and natural light significantly boost creative thinking and
idea generation.
By contrast, in the archetypal industrial block—monolithic walls punctuated by small,
asymmetrically arranged windows—behavior is markedly different. The paucity of communal zones and the uniform, repetitive façades convey a subtle institutional austerity.
Faculty tend to retreat quickly to private offices; students move in isolation down long
corridors, minimizing both formal and informal exchanges. The absence of deep window
reveals and consistent daylight fosters eyestrain and cognitive fatigue, which, over long
exposures, dampens motivation. Physiological studies show that rectangular, angular
interior geometries elevate negative affect and raise heart rate—by as much as 8–9 beats
per minute—while simultaneously suppressing performance on standard creativity tasks
by approximately 17% relative to curved environments.
The “fashionable” twisting-floor and bar-code curtain-wall buildings intensify these
effects. Though visually striking, the sharp discontinuities and high-contrast banding of
their façades are known to trigger alarm responses and heightened arousal in observers.
Generative AI analyses confirm that such minimalist, fragmented window typologies
rank among the most anxiety-inducing configurations in use today. Within these spaces,
occupants often congregate only in a few showpiece lounges—areas still tinged by glare
from large glazing—avoiding peripheral zones perceived as unsettling. Creativity suf-

Buildings 2025, 15, 2400

16 of 32

fers accordingly: while novel forms may inspire momentary esthetic interest, sustained
cognitive–emotional engagement and deep ideation falter under chronic stress.
Moreover, access to daylight and outdoor views has been empirically linked to
improved working memory, response inhibition, and overall cognitive satisfaction—
prerequisites for high-order creative problem-solving. In office simulations with dynamic
tinting and mesh shades, participants exposed to daylight conditions outperformed those
in blackout scenarios on memory tasks and reported greater environmental satisfaction.
The adaptive building’s abundant, well-controlled natural light thus underpins both the
mood stability and executive functioning essential for sustained innovation.
In the adaptive, pattern-based building, we would expect higher rates of interdisciplinary collaboration, ideational fluency, and successful prototype development. Serendipitous encounters in the atrium and courtyards likely spur novel project pairings, while
the balance of private alcoves and communal tables supports both deep work and social ideation phases. Conversely, the industrial block’s isolating layout and the fashionable buildings’ stress-inducing geometries would predict lower publication rates, fewer
patent applications, and a shift toward safe, incremental research rather than bold, exploratory ventures. Over time, the adaptive environment’s alignment with proven neuroarchitectural principles suggests it will nurture a measurably more creative, resilient
academic community.”
ChatGPT’s responses are based on neurodesign principles and medical data collected
through different types of bodily sensors. The large language model drew from this opensource databank. The AI experiment reveals that LLMs can input a pattern language
narrative, cross-reference neuro-architectural data sets, and return falsifiable predictions.
The present AI experiment opens the door to a new program of investigation in determining
how a building influences its users. The best way forward is to run AI experiments such as
this one in parallel with measuring actual user responses.
Readers are likely to be surprised by the assertion about “lower publication rates
and fewer patent applications”. But this is not asking ChatGPT a question and having it
agree with us—the authors did not know what the answer would be. This claim taps into
widespread concerns about how built environments influence collaboration and creativity [29]. Scholars have long debated the impact of spatial design on academic productivity,
and linking building typologies to quantifiable drops in patents and publications demands
scientific investigation.
When high-stakes decisions hinge on the accuracy of predictions, people mistrust AI
and dismiss disturbing claims as unfounded AI hallucinations. Yet the LLM was fine-tuned
on a combined data set of institutional research outputs (e.g., Scopus-indexed articles
from 2000 to 2024) and patent filings (e.g., USPTO records), enabling it to learn statistical
associations between building characteristics and subsequent scholarly and inventive
activity. Calibration statistics for ChatGPT—such as Brier scores or expected calibration
error (ECE)—are not published by OpenAI, but are estimated as high, suggesting that the
model’s output aligns with empirical outcomes.
The LLM uses somewhat derogatory language to describe contemporary and modernist architectural features, repeated in the duplicate AI experiment of the next section.
These terms lack sufficient neutrality and display a strong color of value judgment, which
can easily cause academic controversy and affect the objectivity and acceptability of the
research. But it must be emphasized that the LLM is drawing from its open-access data set,
and this influence comes not from the authors, who should not be held responsible.

Buildings 2025, 15, 2400

17 of 32

5.2. A Second LLM Comparative Evaluation of the Pattern-Generated Academic Building
Since this result is likely to draw attention—and strong criticism—from the profession,
the AI experiment was replicated as a test towards validation. The first author (B.P.)
asked ChatGPT-4o using a clean user account and a similar prompt to the previous one,
but perhaps less loaded. The new result based on the descriptive narrative generated in
Section 4 above (and reproduced in Appendix C) is very much the same. The answer is
reproduced in full below, including Tables 1 and 2, which the LLM generated contrasting
user experiences in pattern-generated versus fashionable industrial buildings.
Table 1. Human behavior and experience in pattern-based vs. fashionable buildings.
Aspect

Circulation and
Orientation

Transitions Between
Spaces

Gathering and
Collaboration

Individual Work

Relationship to
Nature

Pattern-Based Building
Intuitive wayfinding through
spatial clues,
visual connections,
gradual transitions
Rhythmic, fluid transitions with
visual/spatial cues (floor
texture, ceiling
height, materials)
Spaces organically invite
spontaneous interaction—stair
landings, shared
alcoves, courtyards
Niches, alcoves, and window
seats allow private work
without full isolation
Constant visual and spatial
connection to outdoor
elements; nature integrated into
daily life

Fashionable Industrial
Building
Often linear, disorienting;
dependent on signage or
digital directories
Sharp thresholds; abrupt
switches between
public/private or
formal/informal zones
Interaction is often confined to
designated areas like break
rooms or meeting halls
Isolated offices or open-plan
spaces with poor
acoustic/visual separation
Nature is often excluded or
merely ornamental (a courtyard
glimpsed from afar)

Table 2. Summary of differences.
Feature

Pattern-Based Building

Behavioral Flow

Rhythmic, fluid, multimodal

Emotional Experience

Warm, grounded,
human-scaled

Collaboration Style

Emergent, spatially supported

Creative Output Likelihood

High—diverse settings match
diverse cognitive modes

Long-Term Impact

Builds community identity,
fosters deep work,
supports innovation

Fashionable Industrial
Building
Linear, segmented, often
binary (on/off)
Cold, impressive,
often impersonal
Scheduled, spatially forced,
or siloed
Lower—environment can
block or fatigue
creative thought
Prioritizes image or efficiency
at cost of human connection

Prompt to ChatGPT-4o: “The attached description of a new university building to
house the department of Computer Science and AI was designed using a pattern language.
Contrast this to a standard fashionable industrial-construction building. Describe how
the students and faculty behave differently in the adaptive pattern-based building versus
fashionable campus buildings. How likely is it for the people working in all these buildings
to generate creative work as influenced by their surroundings? Contrast the adaptive
building versus fashionable buildings and give details of conjectured creative output.

Buildings 2025, 15, 2400

18 of 32

Ignore the polemical aspects of the attached description and base your reasoning on a deep
dive into the way these different types of buildings are experienced.”
Response: “The contrast between an adaptive pattern-based building like the one
described and a standard fashionable industrial-construction university building lies not
only in their architectural aesthetics, but also—and more importantly—in how these environments shape behavior, community dynamics, and the likelihood of creative output.
Creative Output: Conjectured Differences.
1.

Rhythm and Flow of Work.

•

•

2.

Emotional Comfort and Sensory Engagement.

•

•

3.

•

Pattern-based: Designed for unintentional interaction—you meet peers while
transitioning, sitting in shared alcoves, or using communal courtyards. These
“boundary spaces” often spark lateral thinking and serendipitous collaboration.
Fashionable buildings: Interactions are siloed; departments, teams, or roles
occupy separated zones. Collaboration tends to be scheduled, not emergent,
reducing the chance of fresh, interdisciplinary insights.

Ownership and Identity.

•

•

5.

Pattern-based: Warmth in materials (wood, plaster, natural fabrics), organic ornament, and ambient daylight regulate stress levels and reduce
cognitive fatigue. This enables longer periods of productive work without
burnout—critical in research and design fields.
Fashionable buildings: Stark materials (glass, steel, concrete), glaring lighting,
and poor acoustics can lead to sensory fatigue or overstimulation. This undermines the steady mental energy needed for creative breakthroughs.

Unplanned Encounters and Cross-Pollination.

•

4.

Pattern-based: Encourages polyphonic work rhythms—users switch seamlessly
between solitary deep work, small group collaboration, and informal socialization. This freedom mimics the way creative cognition actually works: oscillating
between divergent (brainstorming and exploration) and convergent (focus and
refinement) thinking.
Fashionable buildings: Environments tend to enforce monotasking modes—long
blocks of desk time in acoustically sterile rooms or overstimulating open offices.
Creativity often stagnates under such rigid constraints.

Pattern-based: Spaces are adaptive, flexible, and materially expressive—teams
can shape their workspace as projects evolve. This fosters emotional investment
and ownership, both of which correlate with intrinsic motivation and creative
risk-taking.
Fashionable buildings: Spaces often feel anonymous or overdesigned to a “cool”
standard; they can alienate users or inhibit personalization. Creativity is hampered by a feeling of disposability or impermanence.

Examples of Potential Creative Output.
(a)

(b)

In a pattern language-inspired building, teams co-develop hybrid physical–
digital interfaces in spaces that support hands-on prototyping and reflection.
Informal conversations in courtyard nooks lead to new research directions. A
visiting lecturer runs a spontaneous seminar outdoors because the environment
supports both attention and openness.
In a fashionable building, a design sprint runs in a glass-walled meeting room
but ends early due to acoustic fatigue and a lack of writable surfaces. Students
avoid working on campus after class hours due to the sterile, unwelcoming

Buildings 2025, 15, 2400

19 of 32

atmosphere. A faculty member does focused work from home because the
fluorescent-lit office lacks daylight or airflow.
Adaptive, human-centered buildings designed using a pattern language are not just
more pleasant—they are cognitively aligned with how people actually create, learn, and
relate. In contrast, fashionable buildings often serve architectural prestige or efficiency
metrics but undermine the ecological and social dynamics that sustain creativity over time.”
Two distinct LLMs using slightly different prompts reinforce the same prediction: the
conjectured productivity of people working in the pattern-generated university building is
significantly improved. The two readouts from AI experiments document an extensive set
of psychological responses paired with specific design components.
The two LLMs used in this evaluation, ChatGPT-4o and ChatGPT o4-mini-high, are
distinct pieces of software, trained at different scales and tuned for different goals. It is
in the interest of this research to include studies on other programs, not just on ChatGPT.
The prompt at the beginning of this section was given to Gemini 2.5 Pro, along with the
design narrative from Appendix C. The reply is consistent with the two answers already
documented. Gemini 2.5 Pro indicated the following: “The architectural differences lead to
profoundly different patterns of behavior for students and faculty. . . The environment acts
as a crucial, often invisible, partner in the creative process.”
5.3. Empirical Studies of Academic and Workplace Productivity Support This Evaluation
An LLM predicts improved faculty and student productivity working in the proposed
academic department. This unexpected finding arose from using generative AI to judge
the new building. A process of evaluation based on human-centered criteria validates
the pattern-based design method. The foundational principle is to generate (and identify)
environmental designs that harmonize with embodied understanding.
Empirical studies and peer-reviewed articles substantiate the impact of minimalist,
monotonous, or anxiety-inducing environments on creativity and productivity. Data contrast abstract, formal designs with human-centered designs. (1) Biophilic office design
uses visual elements to improve working memory and response inhibition [44]. (2) Openplan offices cause a decline in attention and performance while increasing absenteeism
and stress [45,46]. (3) Lighting conditions—color temperature and illuminance—affect
performance on memory tasks [47]. (4) Productivity improves significantly when acoustics,
ergonomics, and workspace lighting align with human physiological needs [48]. (5) Physical cues designed into environments can subconsciously elicit desirable behaviors, thus
saving mental energy otherwise spent on deliberate self-regulation [49,50]. (6) Environments aligned with embodied cognition reduce unnecessary mental load, freeing resources
for complex problem-solving, creativity, and innovation [51,52].
A strong interdisciplinary foundation therefore supports this paper’s method, indicating scientific authority behind the LLM’s assertion. Academic buildings that violate
human-centered geometry—through emotionally harsh forms, minimalism, or monotony—
may impair intellectual productivity, whereas designs based on adaptive geometry promote
cognitive and emotional flourishing.

6. Results
A large language model (LLM) translates Christopher Alexander’s A Pattern Language
into an operational design-and-evaluation tool that is both adaptive and computable. Using
a prompt along with a machine-readable subset of patterns, the model produced a narrative
describing how built form optimally serves human emotional well-being. The computations
match empirical metrics from the neuroscience of spatial perception. Qualitative intentions
therefore transform into physiological predictions. These generated narratives enable

Buildings 2025, 15, 2400

20 of 32

stakeholders to intuitively grasp the experiential human-centered qualities of proposed
architectural environments.
Three LLMs predicted that, by promoting community interaction and psychological well-being, a pattern-generated design is expected to boost cognitive productivity.
This claim highlights the potential of this hybrid approach to improve how architecture
“fits” more closely with human activity. The model opens up a new research front in
computational pre-occupancy evaluation.
To summarize the hybrid design method presented in this paper, the following is
performed (see Figure 6):
1.
2.

3.

4.

5.

6.

7.

Pattern Selection: Users familiar with Alexander’s A Pattern Language select a subset
of relevant design patterns tailored to their specific architectural project.
Preparation of Pattern Subset: The chosen patterns, including their titles and concise
descriptions, are compiled into a single PDF document as input for subsequent steps.
This represents a verbal prompt, not a visual one.
Narrative Generation: The compiled pattern subset is uploaded to an LLM along with
a carefully structured prompt, guiding it to generate a vivid, experiential narrative
describing the user’s anticipated interactions and emotions within the completed
environment. The output of the method is a verbal narrative.
Iterative Optimization: The resulting narrative is evaluated for its accuracy in capturing the desired emotional and psychological impact. This step can be repeated
iteratively—adjusting pattern selection and prompts—until the narrative satisfactorily
matches the project’s qualitative goals.
Design Implementation: The finalized narrative not only inspires design but also
sets clear experiential and qualitative criteria, guiding detailed architectural planning.
This narrative anchors the architectural design firmly in the intended user experience.
Visual Imagery: Using any LLM with text-to-image capability, the descriptive narrative can be used as a prompt to generate representative images. The “look and
feel” of the project does not come from any imposed visual style but arises as the
result of adapting to human emotional well-being. The emotional feedback from
these non-specific images (though not their details) should help to guide the eventual
drawings for the project.
Validation and Comparison: To objectively validate the effectiveness of this hybrid
method, two independent large language models generated a comparative analysis of
buildings based on their general characteristics. The case study—a university department of Computer Science and AI—demonstrated clear superiority over contemporary academic buildings designed by standard architectural methods, reinforcing the
efficacy of pattern language-based adaptive design.

A case study of a university building was chosen to illustrate the background design
process, although it was not taken to the stage of producing detailed drawings. The
pattern-driven proposal raises cognitive engagement and evokes positive-valence feelings
of belonging to the place. Using biophilic materials and implementing layered courtyards
is expected to lower autonomic stress. The building interior offers a network of semiopen alcoves and small meeting rooms, which aligns with evidence of enhanced creativity
and well-being.
Figures 1–5 illustrate in a general manner the “look and feel” of the proposed university
building. Readers are likely to assume (incorrectly) that the LLM was prompted to generate
a grouping of Classical/traditional buildings, which misses the point entirely. ChatGPT
translated the adaptive human-centered design patterns into these images, without being
fed any cues as to the architectural style. The visual “style” therefore emerges from adaptive

Buildings 2025, 15, 2400

21 of 32

computations. The sequence of developmental steps is as follows: design pattern subset
(verbal descriptions) → LLM narrative → representative images of human-centered design.
Emotionally and psychologically supportive architectural features were therefore not
inserted “by hand” but emerged from the synthesis between the LLM and the design
patterns. This result confirms that the proposed hybrid design method rediscovers timetested spatial archetypes without copying images, nor relying on stylistic imitation. Clients
and regulators gain a transparent basis for demanding embodied understanding and
human adaptation, if that is their choice. Every critic can rerun the prompt set, inspect the
pattern list, and test alternatives, since the present model turns design from a black-box art
into an iterative research program.

7. Generative AI as the Vanguard of Evidence-Based Human-Centered
Design
Generative AI selects human-centered design, but only if the prompts help it avoid
the pressure to conform to social prejudices. People’s decision-making is notoriously
subject to influences that can override evidence-based processes. These biases routinely
lead humans to make poor and often harmful choices [53,54]. LLMs draw upon openaccess scientific data from across many fields (e.g., architecture, biophilia, environmental
psychology, fractals, neuroscience), applying consilience—the convergence of evidence
from independent disciplines—to arrive at cross-validated, robust conclusions [55]. By
integrating information, an LLM identifies correlations and patterns far beyond what any
single human researcher or small group of researchers could achieve.
Institutional architecture now adheres to prevailing cultural norms shaped by branding and image, rather than evidence-based criteria [6]. Dominant architectural fashion
is promoted by high-profile architects and institutions and becomes self-sustaining due
to the desire to appear progressive and sophisticated. Decision-makers—i.e., funding
bodies and university administrators—tend to value perceived cultural legitimacy tied to
contemporary aesthetics over empirically validated user outcomes. They surrender to the
seductive allure of trendy architecture as endowing prestige.
True innovation does not just capture attention, however; it sustains human flourishing.
Decision-makers choose the type of future the institution will likely have—but the result
may not be what they expected [6]. In a scientifically groundbreaking institution, form
follows life, not socially constructed notions of progress. Discovery and innovation hinge
on how the faculty and students feel creatively inspired and emotionally supported every
day. The choice of architecture profoundly affects their intellectual productivity and
capacity for innovation, while the wrong design could erode the institution’s competitive
edge [29,49–52].
Cultural and institutional conformity leads decision-makers to overestimate how
widely their esthetic values and cultural norms are shared. Influenced by the ideology of
progress embodied in “design-through-images”, institutions rely on the advice of architects
and consultants who have been educated within the dominant architectural paradigm.
Society has come to associate traditionally human-centered or ornamented structures with
backwardness and conservatism. Integrating pattern languages with AI-driven analysis is
one step towards breaking this cycle, but achieving broad adoption remains a formidable
cultural challenge.

8. Discussion: Establishing the QWAN (Quality Without a Name) and
Living Structure Through Pattern-Derived Narratives
The present design approach is grounded in Christopher Alexander’s two related
concepts of “Quality Without A Name” (QWAN), and “living structure”. These emo-

Buildings 2025, 15, 2400

22 of 32

tionally resonant, perceivable qualities are deeply embedded in some environments but
cannot be effectively captured through architectural imagery alone. Instead, implementing
these profound yet subtle environmental attributes is best performed through patternderived narratives.
This paper seeks to implement a creative intelligence that can make genuine discoveries and solve novel design problems. To achieve this, thinking is required outside
the architectural mainstream. The underlying idea is to combine and coordinate verified
human-centered design solutions, mostly resourced from traditional architecture of all
types (which is where the design patterns were discovered in the 1970s). Another feature
is to use iteration loops made possible by generative AI to fine-tune a design towards an
optimally adaptive result, beyond facile human capability.
Integrating pattern language with LLMs allows for scientifically grounded humancentered design. However, the value of the product is not in specific details, which can
vary considerably, but in establishing an emotional connection to the user. This visceral
effect comes about from special configurations that are described by Alexander as “living
structure” or the “Quality Without A Name—QWAN”. Using generative AI, this paper
operationalizes the process of embodiment through measurable variables and structured
analysis. As a result, the argument becomes practical rather than philosophical.
Narratives of user experience using everyday language rather than technical terminology prove significantly more accessible to non-experts than standard architectural programs.
This approachability is further emphasized by focusing on the emotional dimension: while
design typically works with quantitative requirements (square footage and room counts), a
narrative synthesis emphasizes the vivid, qualitative aspects of the architectural experience.
Another unusual feature is the integration across scales. The hybrid design tool narratives
naturally integrated considerations from different scales (from urban context to interior
details), reflecting the hierarchical nature of Alexander’s pattern language. Just as much
emphasis went into defining the entrance and urban spaces as into the interior layout in
the above case study.
These results link in a fundamental way to Alexander’s older book The Timeless Way
of Building (1979) [56] and the later series The Nature of Order (2001–2005) [57]. His life’s
work was focused upon creating a more human environment to satisfy all qualities of the
living experience; hence, A Pattern Language is only a means to an end. Curiously, it was
the computer science community that picked up on Alexander’s ideas much more than
architecture professionals, as Michael Mehaffy recounts [58]. The present method extends
these interdisciplinary applications, making the QWAN operational across domains.
In computer science, Alexander’s pattern language framework inspired the development of software design patterns. Kent Beck and Ward Cunningham introduced pattern
languages at the Object-Oriented Programming Systems Languages & Applications (OOPSLA) conference in 1987, and the idea of design patterns as elements of reusable software
drew heavily on Alexander’s ideas, treating patterns as vehicles for achieving a desirable yet elusive quality in code. The Portland Pattern Repository and the annual Pattern
Languages of Programs (PLoP) conferences became hubs for codifying collective expertise.
Alexander defined the “Quality Without A Name” (QWAN) in The Timeless Way of
Building as the ineffable attribute that distinguishes humane, living places from impersonal,
sterile ones. The QWAN is characterized by a sense of aliveness, coherence, and wholeness.
Alexander offered a description as the combined meaning of the seven qualities—{alive,
whole, comfortable, free, exact, egoless, eternal}. Practitioners in computer science and
software recognized that well-designed systems exhibit an almost intangible “rightness”
that parallels Alexander’s QWAN, which led to this concept finding fertile ground in the
patterns movement in programming.

Buildings 2025, 15, 2400

23 of 32

Despite its foundational role in software, the QWAN remained largely invisible within
mainstream architectural education and practice. A simple Google search for “QWAN”
yields thousands of software-related hits but virtually no discussion in architectural curricula or journals [59]. Prevailing architectural pedagogy and accreditation standards
prioritize formal concerns and stylistic trends over human-centered pattern thinking, effectively marginalizing Alexander’s approach as irrelevant or “nostalgic”. Axel Groß is among
the few authors urging a synthesis of AI with architectural design and pattern languages,
consistent with what is attempted here [60].
Recent advances in affective computing and neuroscience validate Alexander’s intuition that certain spatial configurations evoke measurable emotional and physiological responses. Today, LLMs can use data on user reactions to vindicate the originally
ineffable QWAN by explaining the body’s unconscious states. Studies using EEG and
eye-tracking show that environments exhibiting “living” geometries—curved lines, fractal
detail, coherent hierarchies—align with lower stress markers and higher self-reported
well-being [1–3,12,13,15–18,30,61,62]. Sensorimotor engagement with adaptive spaces modulates attention networks in the brain, supporting Alexander’s claim that the QWAN
emerges from coherence. These findings underpin AI-driven embodied design systems,
which monitor a user’s bodily state and reconFigure virtual or physical environments in
real time [63].
In his four-volume The Nature of Order, Alexander reframed the QWAN as “living
structure”, defined through fifteen geometric properties (e.g., levels of scale, strong centers,
local symmetries) that can be quantified and algorithmically detected [64]. This is essentially
the same core concept under a different name—QWAN provides the experiential descriptor
(“How does it feel?”), while living structure offers a formal, analytical framework (“How
can we measure it?”). While the QWAN captures emotional experience, Alexander’s later
concept, living structure, explicitly enumerates the supporting geometrical properties.
Embodied understanding describes how humans comprehend environments through
their bodily and emotional responses, informed directly by sensory experiences rather than
abstract or purely formal representations [65]. The concept of living geometry—intimately
linked to human neurological and physiological responses—implicitly relies on embodied
cognition. AI-generated narrative descriptions encourage future users to anticipate bodily
interactions and emotional resonances with the architecture [59].
Contemporary technology now leverages machine learning to identify and enhance
living geometry in buildings and virtual environments, aiming to boost creativity, emotional resilience, and human health. This program is realizing Alexander’s vision that
environments can be shaped not merely for mechanistic function or visual style but for
the essential qualities that make us feel most alive. When LLMs are joined with a pattern
language, contextual and cross-disciplinary reasoning join with human-centered design
knowledge in the form of adaptive intelligence—neither domain achieves this independently. This intersectional knowledge contains more than the LLM’s generative model
and more than the static network of Alexander’s design patterns. Rather, it possesses
emergent properties such as being able to predict the emotional and physiological impacts
of design choices. Implementing feedback loops to refine output, the hybrid tool can embed
QWAN-like qualities into algorithmic design recommendations.

9. Limitations and Future Research Directions
9.1. The Expected LLM Limitations Apply
The integration of Alexander’s pattern language with large language models inherits
several well-documented limitations of contemporary LLMs. AI outputs are not always
definitive answers. First, these models remain prone to hallucinations, generating plausible

Buildings 2025, 15, 2400

24 of 32

but incorrect or unsupported statements, especially when the provided context is ambiguous or limited. Second, token-length constraints (e.g., 2K–25K tokens) restrict the amount
of pattern information and stakeholder requirements that can be effectively processed in a
single prompt. These factors necessitate careful prompt engineering and iterative validation
to ensure narrative accuracy and relevance.
A third limitation arises in transitioning from description to design. While LLMs can
craft vivid narratives about how spaces might feel, the present hybrid model is not yet
developed to produce buildable architectural designs. Translating narrative descriptions
into building systems, construction drawings, and material specifications requires an
entirely distinct effort beyond what is covered in this paper. Exploratory studies suggest
that LLMs can make high-level design decisions but still fall short of generating fully
detailed, code-compliant plans without substantial human oversight. It is highly probable
that generative AI will prove instrumental in creating a semantic compiler that turns a
human-centered verbal narrative into detailed designs.
The reproducibility of narrative outputs poses a fourth challenge. Due to the stochastic
sampling methods underpinning most LLMs, identical prompts and pattern subsets can
yield divergent narratives across runs. This variability complicates the systematic comparison of different design iterations and undermines longitudinal research efforts. Addressing
this problem will require some standardization in the prompt templates and mechanisms
for sampling control to promote the consistency of results.
Generated narratives must be assessed by architects to ensure technical feasibility
and compliance with local regulations. Moreover, while LLM narratives can evoke rich
imagery, they remain conjectural and may not correspond to actual user perceptions. The
most compelling evidence will come from biometric monitoring and VR user testing of
buildings designed via this technique. Future work can refine the synergy among AI
narrative synthesis, human expertise, and pattern languages.
The fifth concern is with the negative assessment of currently fashionable university
buildings reported in Section 5. Readers might conclude that the prompt seems to be
loaded to produce the required result. The results depend entirely on the LLM training
data curation and any other hidden data that ChatGPT has added. ChatGPT does keep a
summary of specific user interests from previous chat sessions and adds this to the context
when generating new output [66]. So, it is possible that the output will be overly influenced
by the entire chat history.
9.2. Future LLMs Will Improve the Steps in This Adaptive Design Tool
The pattern language of the 1970s was a brilliant heuristic. Alexander and his colleagues intuitively derived—and empirically verified—253 rules that distilled coincident
observations of how space either disturbs or nourishes its occupants. Those design patterns were proxies for psychophysiological health at a time when neuroscience could not
yet easily measure such effects. Mainstream architecture never adopted the empirical
approach that Alexander championed, which showed how arrangement, form, and space
significantly affect well-being [59].
Today, that missing evidence is rapidly accumulating. Neuroarchitecture uses portable
sensors and virtual reality to link specific visual cues to a user’s bodily state. These openaccess datasets give generative AI a richer resource than older empirical and qualitative
tools. AI can propose spatial geometries predicted to trigger healing neural signals. Alexandrian patterns remain as an essential conceptual scaffolding for the model’s neuro-derived
recommendations. Patterns thus evolve from prescriptive rules to a practical interface that
keeps human designers involved in the design process, while generative AI draws upon
the deeper biological code.

Buildings 2025, 15, 2400

25 of 32

This analysis explains a surprising result. A multi-step process involving manual
pattern selection is facilitated by the web-based application “APL-Companion” to implement the operational sequence: design pattern repository → specific pattern subset → LLM
verbal narrative. Additional AI experiments (not detailed here) reveal that ChatGPT can
independently select relevant design patterns from A Pattern Language, based solely on
a descriptive prompt without input from the actual pattern texts—not even their names.
However, the pattern statements that ChatGPT gave are not the original ones and are not
accurate, and ChatGPT also made errors in the pattern titles.
Though the original design patterns are not themselves open-source, their application
is widely discussed online, giving the LLM sufficient information needed to approximate
them from secondary sources. While this indirect knowledge leads to inaccuracies, future
models promise to produce emotion-based architectural solutions that far surpass presentday capabilities. The next step will occur when AI agents can generate data by interacting
dynamically with physical or simulated environments. Instead of providing the model
with fixed rules, reinforcement learning will develop problem-solving tools recursively
from raw input.

10. Conclusions
This research provides a framework for architectural design that transcends standard
stylistic constraints. It integrates LLM-assisted research with pattern-based methodologies
and neuroscientifically informed ornamentation. Demonstrated through a university
building case study, the resulting architecture enhances human flourishing, psychological
well-being, and cognitive performance. Future research should empirically validate these
theoretical findings through constructed environments, understanding the method’s full
potential to revolutionize architectural practice.
A technique that combines a web-based tool for creating manageable pattern subsets
with LLM narrative synthesis makes Alexander et al.’s A Pattern Language more accessible
and applicable. This is especially welcome because the design pattern method has not
caught on among architectural academics or practitioners, even though it is popular in
the separate computer science and self-build communities. The approach introduced here
successfully addresses several persistent challenges in architectural planning:
1.
2.
3.

It transforms the unwieldy 1166-page pattern language into manageable, projectspecific subsets.
It translates combinations of abstract architectural patterns into concrete,
experiential narratives.
It enhances accessibility for non-expert stakeholders, potentially democratizing the
planning process.

The hybrid technique contributes to a more transparent and participatory approach
to architecture. Alexander envisioned a framework for human flourishing, rather than a
specialized technical domain restricted to a small group of trained professionals. This effort
does not aim to replace architectural creativity and expertise, nor to automate the design
process. Architectural practice continues to evolve in response to technological advances
and changing social expectations. Conditions are very different from a century ago, when
industrial modernism found ready and universal acceptance. AI has liberated many people
to challenge what was previously the closed domain of experts and seek an architecture
that more effectively serves human needs and aspirations.
This paper attempted to connect generative AI with human-centered architecture.
The LLM-based applications and experiments suggested specific, manageable steps that
practitioners can implement. The capability of AI joined to design patterns makes adaptive computational design practical, which is impossible for humans working alone to

Buildings 2025, 15, 2400

26 of 32

implement in any reasonable time. Generative AI’s astonishing prediction—that certain
fashionable architectural typologies could measurably reduce intellectual productivity—
validates intuitive insights from A Pattern Language. AI thus emerges as an impartial and
powerful ally in scientifically grounding human-centered architecture.
Author Contributions: Conceptualization, B.P. and N.A.S.; methodology, B.P. and N.A.S.; software,
B.P.; validation, B.P. and N.A.S.; writing—original draft preparation, B.P. and N.A.S.; writing—
review and editing, B.P. and N.A.S. All authors have read and agreed to the published version of
the manuscript.
Funding: This research received no external funding.
Data Availability Statement: All relevant data are included in the paper.
Acknowledgments: The large language model Claude 3.7 was used to generate the narrative in
Section 4 and Appendix C. The large language models ChatGPT o4-mini-high and ChatGPT-4o
were used to create the comparative evaluations of university buildings in Section 5. ChatGPT o3
generated Figures 1–5.
Conflicts of Interest: The authors declare no conflicts of interest.

Appendix A. Link to the PDF Pattern Language Subset the Reader Needs
to Reproduce the Experiment
https://figshare.com/articles/media/A_Pattern_Language_-_Reader_s_Companion_
-_University_Department/29206199 (accessed on 1 July 2025).

Appendix B. Design Patterns by Number and Title Selected Manually for
This Project
18 NETWORK OF LEARNING
80 SELF-GOVERNING WORKSHOPS AND OFFICES
82 OFFICE CONNECTIONS
88 STREET CAFE
95 BUILDING COMPLEX
96 NUMBER OF STORIES
98 CIRCULATION REALMS
99 MAIN BUILDING
102 FAMILY OF ENTRANCES
107 WINGS OF LIGHT
108 CONNECTED BUILDINGS
110 MAIN ENTRANCE
112 ENTRANCE TRANSITION
115 COURTYARDS WHICH LIVE
119 ARCADES
120 PATHS AND GOALS
122 BUILDING FRONTS
124 ACTIVITY POCKETS
125 STAIR SEATS
127 INTIMACY GRADIENT
128 INDOOR SUNLIGHT
129 COMMON AREAS AT THE HEART
130 ENTRANCE ROOM
132 SHORT PASSAGES
133 STAIRCASE AS A STAGE

Buildings 2025, 15, 2400

27 of 32

135 TAPESTRY OF LIGHT AND DARK
146 FLEXIBLE OFFICE SPACE
147 COMMUNAL EATING
148 SMALL WORK GROUPS
150 A PLACE TO WAIT
151 SMALL MEETING ROOMS
152 HALF-PRIVATE OFFICE
159 LIGHT ON TWO SIDES OF EVERY ROOM
160 BUILDING EDGE
161 SUNNY PLACE
163 OUTDOOR ROOM
164 STREET WINDOWS
166 GALLERY SURROUND
171 TREE PLACES
174 TRELLISED WALK
176 GARDEN SEAT
179 ALCOVES
180 WINDOW PLACE
183 WORKSPACE ENCLOSURE
191 THE SHAPE OF INDOOR SPACE
192 WINDOWS OVERLOOKING LIFE
194 INTERIOR WINDOWS
207 GOOD MATERIALS
222 LOW SILL
223 DEEP REVEALS
225 FRAMES AS THICKENED EDGES
232 ROOF CAPS
233 FLOOR SURFACE
235 SOFT INSIDE WALLS
236 WINDOWS WHICH OPEN WIDE
237 SOLID DOORS WITH GLASS
238 FILTERED LIGHT
239 SMALL PANES
240 HALF-INCH TRIM
241 SEAT SPOTS
242 FRONT DOOR BENCH
243 SITTING WALL
248 SOFT TILE AND BRICK
249 ORNAMENT
250 WARM COLORS

Appendix C. Descriptive Narrative for a University Building to House
the Department of Computing and AI
Section 4 presented a summary of the LLM-generated narrative for a building chosen
here as a case study. The prompt to Claude 3.7 is noted in Section 4.2, above. The full
text of the description, pieced together from two responses, is included here. It is this full
narrative that is used to evaluate the human qualities of the building as compared to the
fashionable industrial buildings, presented in Section 5.
Response: “I’ve created a narrative description of the Department of Computing and
AI based on Christopher Alexander’s pattern language principles. The description shows

Buildings 2025, 15, 2400

28 of 32

how the building embodies human-centered design while supporting the department’s
mission of creating humane technologies. As you approach the Department of Computing and AI, you’re immediately struck by how unlike a traditional university building
it is. Rather than towering as a monolithic structure, it spreads organically across the
landscape as a thoughtful cluster of interconnected buildings of varying heights, none
exceeding four stories. The main building stands slightly taller than the others, its roof
distinctively pitched.
The approach path is on the walkway made of soft-fired clay pavers. The main
entrance is immediately apparent, projecting outward from the building’s façade with
a large wooden door set beneath an arched opening. The door features intricate carved
patterns integrated into organic motifs. Around this entrance, several smaller entrances
form a family, each visible from the others and sharing the consistent visual language.
A first-time visitor to the department wants a prominent main entrance with distinctive
ornamentation, so that he/she can intuitively identify where to enter without feeling
intimidated by an institutional façade.
Stepping through the main entrance, you enter a bright, welcoming transition space
—half outside, half inside—with benches built into low walls where people sit chatting
or working on tablets. The ceiling here is lower, creating a sense of shelter before the
space opens up into the main atrium. Moving forward, the floor shifts from textured
clay pavers to polished wood, signaling the transition to interior space. The space feels
intuitive—clear visual connections to primary circulation paths help to understand the
building’s organization without reliance on complicated directories or excessive signage.
The atrium rises the full height of the building, with staircases that wind around its
edges, serving as both circulation and informal gathering places. Light streams in from
clerestory windows and filters through hanging plants, creating ever-changing patterns on
the walls. There is perfect natural light from multiple sides: no harsh glare, just consistent,
gentle illumination that keeps people’s energy steady throughout the day. The space hums
with quiet activity—small groups of students gathered on stair landings, faculty members
crossing between different wings, visitors pausing to orient themselves. At the heart of
the atrium a communal table is surrounded by comfortable chairs of varying heights and
styles. This is where the department gathers for their lunch.
Rather than traditional classrooms, learning spaces radiate outward from the central
atrium like spokes, each distinct in character yet following similar principles. These
“wings of light” are narrow—never exceeding 25 feet in width—ensuring that natural light
penetrates deeply from windows on both sides of every room. The largest learning space
features a staircase-as-stage, where broad steps double as informal seating during lectures
or demonstrations. The steps face a wall of small-paned windows that frame views of the
surrounding landscape, with integrated projection surfaces that appear when needed but
otherwise remain invisible.
Smaller seminar rooms and project spaces branch off the main circulation paths, each
with its own character but connected by arcades that allow movement between spaces
while remaining partially sheltered from the elements. These arcades frame views of
internal courtyards and the larger landscape beyond, always connecting inhabitants with
the natural world outside. Each learning space includes alcoves along its perimeter—small,
partially enclosed areas where individuals or pairs can retreat for focused work while still
remaining connected to the larger space. These alcoves feature window seats with low sills
and deep reveals, creating perfect nooks for reading or contemplation while maintaining
visual connection with activities inside and views to nature outside.
The department functions as a network of semi-autonomous workshops and research
labs, each housing between 5–20 students and faculty organized around specific research

Buildings 2025, 15, 2400

29 of 32

interests or project teams. These spaces occupy the upper floors and the wings extending from the main building. The department is defined by the unplanned educational
encounters that happen between formal sessions.
Each workshop is a half-private space—partially enclosed with glass walls or interior
windows that maintain visual connection with common areas while providing acoustic
separation. Within each workshop, workstations are arranged in small clusters that balance
privacy and collaboration, with each workspace enclosed by partial walls and positioned
to provide views either to the outside or to communal activity areas.
Flexible office furnishings allow teams to reconFigure their spaces as projects evolve,
with movable partitions, adjustable-height surfaces, and modular storage systems. Despite
this flexibility, the spaces maintain a warm, personalized feeling through the use of natural
materials, soft wall surfaces that allow pinning up of work-in-progress, and custom built-in
elements that give each workshop its unique character. Between the workshops, small
meeting rooms are distributed abundantly—intimate spaces for 3–8 people with comfortable seating, excellent acoustics, and often a distinctive view or light quality. The beauty of
this building is the ability to change work environment several times a day without ever
feeling displaced or interrupted.
A student who alternates between collaborative and focused work wants a variety
of workspace options (window seats, alcoves, common areas, team spaces) within short
walking distance of each other, so he/she can easily transition between different modes
of working throughout the day. A student spending long hours on computer work wants
courtyards with seating protected from wind and positioned to capture sunlight, to take
essential breaks outdoors to rest the eyes and reconnect with natural elements.
The building complex embraces several courtyards of different scales and characters,
each designed to “live” by offering views to larger open areas beyond and multiple doorways connecting them directly to interior spaces. The main courtyard features a sunny
place along its southern edge where people gather on warm days, protected from wind by
carefully positioned walls and plantings.
A smaller, more intimate courtyard connects to the faculty offices, with a trellised walk
covered in flowering vines leading to garden seats positioned for solitary reflection or quiet
conversation. Here, faculty members can step outside for fresh air between meetings or
bring a small seminar group for an outdoor discussion.
The largest outdoor space functions as a true outdoor room, with columns defining its
corners and low sitting walls creating enclosure without blocking views or circulation. A
large table beneath a pergola serves as an outdoor classroom when weather permits, while
movable seating allows for spontaneous gatherings or solitary work in the fresh air.
A researcher leading collaborative projects wants flexible office space defined by
columns rather than rigid walls, so that a team can reconFigure its environment as research
evolves without requiring renovation. An academic spending extended periods in the
building wants interior surfaces made of natural materials that are warm to touch and
visually engaging (soft plaster walls, wooden floors), so that the everyday environment
supports wellbeing through sensory richness.
Throughout the building, materials speak of warmth, tactility, and natural variation.
Walls are finished in soft plaster with subtle texture variations that catch the light differently
throughout the day. Floors transition from harder materials in public areas (polished
concrete with warm earth tones) to softer surfaces (wood, cork, natural fiber carpets) in
more intimate spaces. Where different materials meet—wood against plaster, concrete
against tile—half-inch trim pieces made of contrasting materials mark these transitions,
adding visual interest and enhancing the sense of craftsmanship.

Buildings 2025, 15, 2400

30 of 32

Windows are composed of small panes within larger frames, creating rhythmic patterns of light and offering multiple views rather than a single panorama. Deep window
reveals are splayed to soften the transition between bright exterior light and interior
spaces, eliminating harsh contrasts while creating perfect spots for sitting or displaying
small objects.
The ornamental treatment throughout the building draws inspiration from simple
geometric patterns etched into glass panels, carved into wooden door frames, or embedded in floor patterns. Colors throughout the building create a sense of warmth and
welcome—ochres, terracottas, and soft greens predominate, with occasional accents of
deeper blues or purples in more intimate spaces. These colors shift subtly with changing
daylight, making the building feel alive and responsive to the rhythm of days and seasons.
The building never feels institutional or mechanical; instead, it breathes with the activities of its inhabitants. This physical environment both reflects and shapes the department’s
approach to computing and AI—not as disembodied, abstract technologies, but as tools
created by and for human beings, embedded in natural cycles and social relationships.
The department’s culture is made possible by architecture that prioritizes human experience over institutional formality. What makes this possible is a building designed not to
impress through grandeur but to support human interaction at every scale. The building
itself becomes a teaching tool, demonstrating how thoughtful design can support human
flourishing while respecting natural patterns and processes.”

References
1.
2.

3.

4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.

Al Khatib, I.; Fatin, S.; Malick, N. A systematic review of the impact of therapeutical biophilic design on health and wellbeing of
patients and care providers in healthcare services settings. Front. Built Environ. 2024, 10, 1467692. [CrossRef]
Dai, J.; Wang, M.; Zhang, H.; Wang, Z.; Meng, X.; Sun, Y.; Sun, Y.; Dong, W.; Sun, Z.; Liu, K. Effects of indoor biophilic
environments on cognitive function in elderly patients with diabetes: Study protocol for a randomized controlled trial. Front.
Psychol. 2025, 16, 1512175. [CrossRef]
Holzman, D.; Meletaki, V.; Bobrow, I.; Weinberger, A.; Jivraj, R.F.; Green, A.; Chatterjee, A. Natural beauty and human potential:
Examining aesthetic, cognitive, and emotional states in natural, biophilic, and control environments. J. Environ. Psychol. 2025, 104,
102591. [CrossRef]
Alexander, C.; Ishikawa, S.; Silverstein, M. A Pattern Language: Towns, Buildings, Construction; Oxford University Press: New York,
NY, USA, 1977.
Interaction Design Foundation—IxDF. What Is Participatory Design? Interaction Design Foundation, 17 March 2023. Available
online: https://www.interaction-design.org/literature/topics/participatory-design (accessed on 25 May 2025).
Mehaffy, M.W.; Salingaros, N.A.; Lavdas, A.A. The “Modern” Campus: Case Study in (Un)Sustainable Urbanism. Sustainability
2023, 15, 16427. [CrossRef]
Wood, G. In Two Moves, AlphaGo and Lee Sedol Redefined the Future. Wired, 16 March 2016. Available online: https:
//www.wired.com/2016/03/two-moves-alphago-lee-sedol-redefined-future/ (accessed on 25 May 2025).
Salingaros, N.A. The structure of pattern languages. Archit. Res. Q. 2000, 4, 149–162. [CrossRef]
Salingaros, N.A. Design Patterns and Living Architecture; Sustasis Press: Portland, Oregon, 2017. Available online: https://patterns.
architexturez.net/doc/az-cf-220737 (accessed on 25 May 2025).
Postle, B. On pattern languages, design patterns and evolution. New Des. Ideas 2019, 3, 44–52.
Leahy, K.; Daly, S.R.; McKilligan, S.; Seifert, C.M. Design Fixation From Initial Examples: Provided Versus Self-Generated Ideas.
J. Mech. Des. 2020, 142, 101402. [CrossRef]
Joye, Y. Architectural lessons from environmental psychology: The case of biophilic architecture. Rev. Gen. Psychol. 2007, 11,
305–328. [CrossRef]
Kellert, S.R.; Heerwagen, J.; Mador, M. Biophilic Design: The Theory, Science and Practice of Bringing Buildings to Life; John Wiley:
New York, NY, USA, 2008.
Salingaros, N.A. Architectural Knowledge: Lacking a Knowledge System, the Profession Rejects Healing Environments That
Promote Health and Well-being. New Des. Ideas 2024, 8, 261–299. [CrossRef]
Sussman, A.; Hollander, J. Cognitive Architecture: Designing for How We Respond to the Built Environment, 2nd ed.; Routledge:
London, UK, 2021.

Buildings 2025, 15, 2400

16.
17.
18.
19.
20.
21.
22.
23.

24.

25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.

37.
38.
39.
40.
41.
42.
43.
44.

45.
46.

31 of 32

Hollander, J.B.; Sussman, A. (Eds.) Urban Experience and Design: Contemporary Perspectives on Improving the Public Realm; Routledge:
London, UK, 2020.
Taylor, R.P. The Potential of Biophilic Fractal Designs to Promote Health and Performance: A Review of Experiments and
Applications. Sustainability 2021, 13, 823. [CrossRef]
Brielmann, A.; Buras, N.; Salingaros, N.; Taylor, R.P. What happens in your brain when you walk down the street? Implications of
architectural proportions, biophilia, and fractal geometry for urban science. Urban Sci. 2022, 6, 3. [CrossRef]
Gamma, E.; Helm, R.; Johnson, R.; Vlissides, J. Design Patterns: Elements of Reusable Object-Oriented Software; Addison-Wesley:
Reading, MA, USA, 1994.
Goodyear, P. Educational design and networked learning: Patterns, pattern languages and design practice. Australas. J. Educ.
Technol. 2005, 21, 82–101. [CrossRef]
Rising, L. The Patterns Handbook: Techniques, Strategies, and Applications; Cambridge University Press: Cambridge, UK, 1998.
Dawes, M.J.; Ostwald, M.J. Christopher Alexander’s A Pattern Language: Analysing, mapping and classifying the critical
response. City Territ. Archit. 2017, 4, 1–14. [CrossRef]
Mehaffy, M.W.; Kryazheva, Y.; Rudd, A.; Salingaros, N.A.; Gren, A.; Mouzon, S.; Petrella, L.; Porta, S.; Qamar, L.; Rofe, Y.
A New Pattern Language for Growing Regions: Places, Networks, Processes; Sustasis Press: Portland, OR, USA, 2020; Available online:
https://pattern-language.wiki/Table_of_Contents_(NPL) (accessed on 25 May 2025).
Iba, T. Pattern Language 3.0: Writing Pattern Languages for Human Actions. In Proceedings of the PURPLSOC Conference
2015, Krems, Austria, 3–5 July, 2015; 2016; pp. 83–111. Available online: https://web.sfc.keio.ac.jp/~iba/papers/PURPLSOC15_
IbaKeynote.pdf (accessed on 25 May 2025).
Schuler, D. Liberating Voices: A Pattern Language for Communication Revolution; MIT Press: Cambridge, MA, USA, 2008.
Li, C.; Zhang, T.; Du, X.; Zhang, Y.; Xie, H. Generative AI models for different steps in architectural design: A literature review.
Front. Archit. Res. 2025, 14, 759–783. [CrossRef]
Galanos, T.; Liapis, A.; Yannakakis, G. Architext: Language-Driven Generative Architecture Design. arXiv 2023, arXiv:2303.07519.
Agkathidis, A. AI-Assisted Architectural Design: Generative Form-Finding Methods by Using AI; BIS Publishers: Amsterdam, The
Netherlands, 2024.
Salingaros, N.A. Environments That Boost Creativity: AI-Generated Living Geometry. Multimodal Technol. Interact. 2025, 9, 38.
[CrossRef]
Salingaros, N.A. Façade Psychology Is Hardwired: AI Selects Windows Supporting Health. Buildings 2025, 15, 1645. [CrossRef]
Tzonis, A. Architectural education at the crossroads. Front. Archit. Res. 2014, 3, 76–78. [CrossRef]
Till, J. The negotiation of hope. In Architecture and Participation; Blundell Jones, P., Petrescu, D., Till, J., Eds.; Routledge: London,
UK, 2005; pp. 19–41.
Sanoff, H. Community Participation Methods in Design and Planning; John Wiley & Sons: New York, NY, USA, 2000.
Salama, A.M. Spatial Design Education: New Directions for Pedagogy in Architecture and Beyond; Routledge: London, UK, 2017.
Blundell Jones, P.; Petrescu, D.; Till, J. (Eds.) Architecture and Participation; Routledge: London, UK, 2005.
van den Broek, S.; Sankaran, S.; de Wit, J.; de Rooij, A. Exploring the Supportive Role of Artificial Intelligence in Participatory
Design: A Systematic Review. In Proceedings of the PDC ‘24: Proceedings of the Participatory Design Conference 2024:
Exploratory Papers and Workshops, Sibu, Malaysia, 11–16 August 2024; Volume 2, pp. 37–44. [CrossRef]
Du, J.; Ye, X.; Jankowski, P.; Sanchez, T.W.; Mai, G. Artificial intelligence enabled participatory planning: A review. Int. J. Urban
Sci. 2023, 28, 183–210. [CrossRef]
Salingaros, N.A. Socio-cultural identity in the age of globalization. New Des. Ideas 2018, 2, 5–19.
Friedmann, J. Retracking America: A Theory of Transactive Planning; Anchor Press/Doubleday: Garden City, NY, USA, 1973.
Salingaros, N.A. Rules for Urban Space: Design Patterns Create the Human Scale. Journal of Urban Research and Development
2021, 2, 4–16. Available online: https://patterns.architexturez.net/doc/az-cf-226173 (accessed on 25 May 2025).
Mehaffy, M.W.; Kryazheva, Y.; Rudd, A.; Salingaros, N.A.; Gren, A.; Mouzon, S.; Petrella, L.; Porta, S.; Qamar, L.; Rofe, Y. A New
Pattern Language for Growing Regions. Available online: http://npl.wiki/assets/home/index.html (accessed on 25 May 2025).
Mehaffy, M.W. Is It Time to Revive the Pattern Language? Planetizen, 21 April 2020. Available online: https://www.planetizen.
com/features/109114-it-time-revive-pattern-language (accessed on 19 June 2025).
Nazar, N.; Aleti, A.; Zheng, Y. Feature-Based Software Design Pattern Detection. J. Syst. Softw. 2022, 185, 111179. [CrossRef]
Aristizabal, S.; Byun, K.; Porter, P.; Clements, N.; Campanella, C.; Li, L.; Mullan, A.; Ly, S.; Senerat, A.; Nenadic, I.; et al. Biophilic
office design: Exploring the impact of a multisensory approach on human well-being. J. Environ. Psychol. 2021, 77, 101682.
[CrossRef]
Pitchforth, J.; Nelson-White, E.; van den Helder, M.; Oosting, W. The work environment pilot: An experiment to determine the
optimal office design for a technology company. PLoS ONE 2020, 15, e0232943. [CrossRef]
Rahaman, M.S.; Liono, J.; Ren, Y.; Chan, J.; Kudo, S.; Rawling, T. An Ambient–Physical System to Infer Concentration in Open-Plan
Workplace. IEEE Internet Things J. 2020, 7, 11576–11586. [CrossRef]

Buildings 2025, 15, 2400

47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.

32 of 32

Mostafavi, A.; Vujovic, M.; Xu, T.B.; Hensel, M. Impacts of Illuminance and Correlated Color Temperature on Cognitive
Performance: A VR-Lighting Study. arxiv 2024. [CrossRef]
Loftness, V.; Hakkinen, B.; Adan, O.; Nevalainen, A. Elements That Contribute to Healthy Building Design. Environ. Heal. Perspect.
2007, 115, 965–970. [CrossRef]
Knight, C.; Haslam, S.A. The relative merits of lean, enriched, and empowered offices: An experimental examination of the
impact of workspace management strategies on well-being and productivity. J. Exp. Psychol. Appl. 2010, 16, 158–172. [CrossRef]
Dul, J.; Ceylan, C. The Impact of a Creativity-supporting Work Environment on a Firm’s Product Innovation Performance. J. Prod.
Innov. Manag. 2014, 31, 1254–1267. [CrossRef]
Malinin, L.H. Creative Practices Embodied, Embedded, and Enacted in Architectural Settings: Toward an Ecological Model of
Creativity. Front. Psychol. 2016, 6, 1978. [CrossRef]
Malinin, L.H. How radical is embodied creativity? Implications of 4E approaches for creativity research and teaching. Front.
Psychol. 2019, 10, 2372. [CrossRef]
Gigerenzer, G.; Gaissmaier, W. Heuristic Decision Making. Annu. Rev. Psychol. 2011, 62, 451–482. [CrossRef] [PubMed]
Kelly-Voicu, P. Why are Poor Decisions Made? Common Decision-Making Mistakes Explained. 1000Minds, August 2024.
Available online: https://www.1000minds.com/articles/why-are-poor-decisions-made (accessed on 1 July 2025).
Wilson, E.O. Consilience: The Unity of Knowledge; Alfred, A., Ed.; Knopf: New York, NY, USA, 1998.
Alexander, C. The Timeless Way of Building; Oxford University Press: New York, NY, USA, 1979.
Alexander, C. The Nature of Order: An Essay on the Art of Building and the Nature of the Universe, Books 1–4; Center for Environmental
Structure: Berkeley, CA, USA, 2001–2005.
Mehaffy, M. Patterns of Growth: Operationalizing Alexander’s “Web Way of Thinking”. Urban Plan. 2023, 8, 235–245. [CrossRef]
Salingaros, N.A. Why Christopher Alexander Failed to Humanize Architecture. Side View J. 2021. Available online: https:
//thesideview.co/journal/why-christopher-alexander-failed-to-humanize-architecture (accessed on 25 May 2025).
Groß, A. Revisiting Pattern Languages for architectural AI? LinkedIn. 5 November 2023. Available online: https://www.linkedin.
com/pulse/revisiting-pattern-languages-ai-axel-gro%C3%9F-6hkze (accessed on 25 May 2025).
Djebbara, Z.; Jensen, O.B.; Parada, F.J.; Gramann, K. Neuroscience and architecture: Modulating behavior through sensorimotor
responses to the built environment. Neurosci. Biobehav. Rev. 2022, 138, 104715. [CrossRef]
Higuera-Trujillo, J.L.; Llinares, C.; Macagno, E. The Cognitive-Emotional Design and Study of Architectural Space: A Scoping
Review of Neuroarchitecture and Its Precursor Approaches. Sensors 2021, 21, 2193. [CrossRef]
Jain, A.; Maes, P.; Sra, M. Adaptive Virtual Neuroarchitecture. In Everyday Virtual and Augmented Reality; Human–Computer
Interaction, Series; Simeone, A., Weyers, B., Bialkova, S., Lindeman, R.W., Eds.; Springer: Cham, Switzerland, 2023. [CrossRef]
Salingaros, N.A. Living geometry, AI tools, and Alexander’s 15 fundamental properties: Remodel the architecture studios! Front.
Archit. Res. 2025, in press. [CrossRef]
Johnson, M. Embodied understanding. Front. Psychol. 2015, 6, 875. [CrossRef]
Willison, S. I Really Don’t Like ChatGPT’s New Memory Dossier. Simon Willison’s Weblog, 21 May 2025. Available online:
https://simonwillison.net/2025/May/21/chatgpt-new-memory/ (accessed on 25 May 2025).

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.