mahnoor·fatima

Captain CAT

The Captain CAT application on a laptop. Coloured nodes labelled play, tree, flower, rest, pond, path and sit are connected by thin lines inside a dashed rectangle marked COURTYARD BOUNDARY, with the resulting coloured courtyard plan alongside it.

A climate-aware design co-pilot for courtyards that think, adapt, and respond.

Urban courtyards are overheating, underused, and disconnected from how people actually live. Captain CAT reimagines them as intelligent, social, climate-responsive spaces through conversation, data, and design logic.

Pixel-art orange tabby cat walking in profile.

Origin Story

Courtyards sit at the intersection of climate, community, and daily life, yet they are often designed through static diagrams and late-stage simulations. Captain CAT was developed to shift this moment earlier, transforming vague intentions into spatial logic, climate feedback, and geometry in real time. Acting as a friendly design co-pilot, CAT invites dialogue instead of prescriptions, making climate intelligence accessible during concept design.

The tool in use, from a written brief to a resolved courtyard.
Screen recording of the Captain CAT application. Its Climate Analysis tab returns quick analysis results for Austin, United States, alongside a Rhino viewport showing the coloured courtyard plan.
The tool running a climate analysis for Austin, with the courtyard plan updating alongside it in Rhino.

How Does CAT Work?

Workflow diagram as a chain of circled icons joined by dotted lines, running from start through data, design board, graph-based configuration, graph query and image generation to a climate based activity generator and the output.
What this diagram says

The pipeline is labelled in four bands. Inputs: start, data, case studies, design parameters, ML data, climate data, design board, user requirements. Graph stage: graph query, graph-based configuration, interactive graph. Geometry: possible geometries, courtyard division, multiple versions. Analysis and output: comfort analysis, climate based activity generator, activity suggestions, image generation, output.

Conversational Input

Designers describe needs in natural language, and CAT parses them into programmatic intents, constraints, and adjacencies instead of fixed forms.

  • Shade
  • Play
  • Water
  • Rest
  • Visibility

Graph Based Learning

Each courtyard function becomes a node in a custom graph system. Relationships encode:

  • Proximity. What wants to sit near what.
  • Conflict. What cannot share an edge.
  • Synergy. What gets better together.

Anchor-driven clustering then relaxes the graph across a surface, generating legible spatial distributions rather than arbitrary layouts.

Climate Intelligence

Environmental data enters early. EPW climate files and UTCI analysis inform where people can linger, not just where space exists. Shade, radiation, and comfort are evaluated live as the layout evolves.

Parametric Translation

The graph resolves into geometry inside Grasshopper and Rhino. Designers see immediate 3D feedback, while CAT critiques:

  • Connectivity
  • Clustering quality
  • Climate performance
Axonometric view of a courtyard divided into flat polygonal zones in shades of green and yellow, each labelled with its function, with simplified tree canopies placed over them.
The graph resolved into courtyard geometry, one zone per function.
Softly rendered courtyard plan generated from the tool's output. A green lawn is ringed with pink, yellow and blue flowering planting around a pale blue pool, set inside the tan footprint of the surrounding building.
An SDXL pass over the same output, turning a zoning diagram into something a client can read.

Generative Imagination

Screenshots are processed through SDXL to generate plans and concept visuals, bridging analysis and storytelling. Each iteration is archived for clients and collaborators as structured PDFs, CSVs, and JSONs.

Graph ML Insights

Courtyard and building elements were transformed into a Neo4j graph containing spaces, windows, adjacencies, and environmental metrics. Community detection revealed hidden spatial patterns, informing targeted design improvements.

Three-dimensional scatter plot on labelled x, y and z axes. Several hundred nodes in black, blue and orange are linked by grey edges into a dense sphere.
The Neo4j graph of spaces, windows, adjacencies and environmental metrics, plotted for community detection.

Real-Time Comfort Prediction

A trained Random Forest model predicts thermal comfort instantly inside Grasshopper, replacing slow simulation cycles with immediate feedback.

  • 5,000+ courtyard configurations generated and simulated using UTCI
  • Instant thermal comfort prediction, in place of a simulation cycle
A dense grid of hundreds of small axonometric thumbnails, each a grey courtyard block with a different arrangement of green massing inside it, every one captioned with its parameter string.
Part of the training set. Each thumbnail is one simulated configuration, captioned with the parameters that produced it.

Captain CAT is a design intelligence layer that closes the loop between intent, environment, and form. By treating courtyards as living systems rather than leftover space, the project demonstrates how AI can augment architectural judgment without replacing it.

This framework scales beyond courtyards to streets, plazas, and adaptive public spaces, pointing toward climate-responsive digital twins that learn alongside designers.

Project developed in collaboration with Andrea Ardizzi, Leila Sheikh, and Lennart Hamm for IAAC. You can read the original article on the IAAC blog.