More on AI-Enabled Modeling

[Mark Vriesenga (YDKDY)]

Here is an unedited video of an ai-enabled modeling session:

1) Load a document corpus

2) Select models to make

3) Use intent-based chat to interact

4) Documentation

5) Settings

This is emerging as the future of MBSE.  The good news is that creating a StarUML plugin is easy.  This plugin provides a web browser (to connect to the web app backend) and a web socket to expose a programmable API.  The websocket essentially allows the backend webapp to program StarUML via remote control.

Enjoy the video.

https://limewire.com/d/9kmdn#hACa8xRDLZ

[Gregor Reich]

Hi, thanks for sharing your insights — this is indeed very interesting!

Unless it's confidential, can you share a bit more detail on what the input was in your example (video)?

Thank you and best regards, Gregor.

[Mark Vriesenga (YDKDY)]

Of course.  The input is a test data set I created.  They emulate an Operations Concept, a Concept of Operations, and a set of Mission/User Scenarios.  The files are narrative text.  See attached files.

The ingest used standard retrieval augmentation generation (RAG) techniques: cleaning the document, chunking the document, and vectorizing (adding a feature vector) to support search.

The diagram extractors process each document chunk to look for model elements and relations.  An LLM is used for the discovery process.  

The plugin with the web browser and exposed StarUML API allows the external code to draw the diagrams.  There is much room for improvement here.  

The chat client uses natural language processing (NLP) and intent-based routing to execute commands.  I did not show all the intent-based routing functions in the video.  One that is particularly useful is generating documentation for model elements from the ingested document set.

This is where AI-Enabled MBSE is going.  I am hoping to get this community thinking about this, as I have always liked StarUML as my 'quick' modeling tool.

Mark