Module 8 Questions:
I chose Option A. The main problem the speakers are addressing is that building performance analysis is most useful early in the design process, when changes are still easy and inexpensive, but traditional simulation workflows are often too slow and technical to fit naturally into early design exploration. Their proposed solution is to generate synthetic building-performance data using Dynamo, Generative Design in Revit, and Autodesk Insight, and use that data to train machine-learning surrogate models that can provide faster analysis feedback.
One thing that surprised me was the lack of usable data in the AEC field. As buildings are responsible for such a large share of carbon emissions, I expected there to already be large and organized datasets of building models and performance results. Instead, the talk shows that the data is often incomplete, inconsistent, not diverse enough, or difficult to access.
What I would be interested in trying is connecting this idea to what I am currently learning in my Engineering Design Optimization course. I would like to explore different ways of creating and training surrogate models, then compare their predictions to real simulation or optimization results. It would be especially interesting to see which type of model is most reliable depending on the design context, for example for buildings controlled by wind, seismic performance, energy use, or other constraints.
During the quarter, I felt that the most challenging part was debugging and troubleshooting the workflows, whether it was code that was not working or results that seemed wrong. The software we used often gave very short or unclear error messages, so it was sometimes difficult to know where the actual problem was. I think a built-in AI tool that could analyze the graph, give warnings, explain possible issues, and act like a side chat would be extremely helpful. This would be very useful when working with long graphs with many nodes and connections, where it is not always obvious whether the final result is correct or whether the graph is doing exactly what it is supposed to do. At the same time, I think debugging the graph ourselves is an important part of the learning process, because this is where a lot of the understanding of the software actually happens. Still, an AI tool would definitely make the process more efficient without completely replacing the learning.
Raven : Raven is an AI co-pilot for Grasshopper that helps users create geometry, set up workflows, and automate repetitive modeling tasks by using written prompts. This would have helped me this quarter because a lot of the work involved building and debugging long graphs. Having a specialized tool that could explain the logic, suggest nodes, or help find mistakes would have made the process faster.
Floor Plan AI : It helps generate and edit floor plans quickly from basic inputs or prompts (given the # of rooms, bathrooms, areas…) . This would not have changed much in my work this quarter, but I think it could be useful later. As a structural engineer, I am not an expert in architectural plans, so if a simple change is needed, a tool like this could help me test it quickly instead of sending plans back and forth and waiting each time.
Trunk Tools : It uses AI to help construction teams search through and organize project information such as drawings, documents, and schedules. I think this is very useful for the built environment because construction is very complicated and unpredictable, and projects often contain a huge amount of unorganized information. A tool like this could help people find answers faster and reduce confusion during the construction process, making it a more efficient process.