But media research failed again and turned out poor imagery with a few interesting things.
And then... the script of the following episode tried to use the same article on black holes. I then realized that the knowledge Graph wasn't going to track this unless I managed to connect the news archive with the finished, aired scripts in a different collection.
I spent the last couple of days rebuilding the Graph logic, and after a few hopeless nights, it's working well: storing articles and full episodes with metadata and temporal tracking, and finally understanding how assets from different collections are related.
This enabled the analyst agent to understand which stories were previously covered and decide if there are new information to add or follow up. If not, it replaces with a new article.
It’s exciting to see a simple idea working, and I was able to set up a new tool for extending the knowledge graph to other collections.
New episode coming out tomorrow!