This project was created at the 2018 Thornton Tomasetti Hackathon in NY. Dynamo node suggestion based on user history using Machine Learning.

Project GitHub

Dynamo 2.0 (JSON)
Revit 2019
Dynamo View Extension

Accord - Markov Chain, Naive Bayes Classifier
ML.NET - Fast Tree Regression
Lunchbox ML - Prototyping Environment

Final Demo Live Dynamo node autocomplete as graph is built (post-training)


For the POC application in the hackathon, we used Markov Chains and a Naive Bayes Classifier. Although that achieved most of the results we were looking for, we decided that next steps would be to look at models which learn “live”, and look at deep learning models which may be able to give more accurate predictions based on the greater context of the definition, rather than a small piece of what the user is doing.

First Demo Basic node & graph autocomplete (post-training).


The final application was deployed as stand-alone add-in, and as a part of the Dynamo Package manager. Users were allowed to user their own Dynamo definitions for model training, or going with the pre-trained models from the collection of definitions provided by Konrad Sobon and Petr Mitev.

Package Manager Debut on the Dynamo Package Manager.

Reception Public excitement around the release!