Sheel Sansare - Part 1

Intended users

The target audience is designers who use Revit, particularly those who are new to Revit.

Need you’re trying to provide a solution or support for

Revit is very crowded with buttons and features that can get overwhelming for both new designers and existing users. As the number of features continues to grow, the learning curve gets steeper. With advances in AI through Large Language Models (LLM), there is finally room for innovation within Revit automation. I propose a natural language interface for Revit.

Inputs

My tool has two very simple inputs. A Revit component, and a natural language command in the form of a string. For example, the user can choose a parametric tower and say “change the building height to 100ft”.

Underlying logic of the model you’ll implement

I will use the OpenAI api (that also powers ChatGPT) to implement this task. Given the string input, I will prompt the LLM to determine which parameter the user is hoping to change. Similarly I will prompt the LLM to extract the number from the input. With the extracted parameter name and number, I will simply set the element parameter value using a dynamo node.

Outputs

The output is a modified Revit component in the Revit interface.