In 2018 I successfully loaded Rasa NLU on my home server. I trained it with different input and it was able to determine the appropriate intent about 80% of the time. The intents that I used were
The affirm, goodbye, greet and negate intents would respond with a simple one or two word reply. The others, except for search, would simply confirm the intent categorization. The search intent was programmed to actually run a google search and respond with the first result.
I integrated a slack bot that I named Otto that could chat with my Rasa NLU code using the Slack web interface. The search results that I received were anything from links to images to videos. It was ok. There is definitely much more to explore with Artificial Intelligence.
Here are some samples of interacting with Otto via Slack.
In this conversation, I asked Otto if he was still running in the background. The search results responded with an Overwatch forum post about the process still running in the background.
On June 26th, 2018 I tried telling Otto to say goodbye first and it responded with search results for spanishdict.com
Then I wanted to know the score of the Dodgers game. Unfortunately, the search results were for games played on April 10th, 2017 instead of the date I was searching, June 26th, 2018. Needless to say there’s some work to be done to improve the program.