In this demo, I would like to demonstrate the implementation of Slackbot based on GoodData enriched by ChatGPT. The purpose of this demo is to show and motivate how the integration with ChatGPT can be done. Even though, ChatGPT is not 100% percent ready for production and not even for integration.
Disclaimer:
This demo is only for research. Please do not use it in production, and beware of the licenses of used libraries.
Big thanks go to Sergey Levchenko, and Jan Panský who worked with me on the base PoC for this demo.
If you'd like to run this demo, please follow the instructions in INSTRUCTIONS.md.
You can find the pure integration of Slack with ChatGPT in the pure_integration.py file. The implementation is simple, and it takes the text sent to the Slackbot, passes the text to ChatGPT, and then the response is sent back to Slack by Slackbot.
Beware that you can use text formatting in Slack, such as bold, italic, code, etc. These changes will also be sent to ChatGPT.
Unfortunately, at the time this demo was created, there were no ways to teach ChatGPT new things persistently. Because of that, we need to do some hacks or, let's say, preprocessing of the information to teach ChatGPT something new. Let's have a look.
The use of metadata to teach ChatGPT about GoodData is presented in metadata.py. The idea behind the implementation is to give ChatGPT information about GoodData first and then use the knowledge of ChatGPT.
Warning ChatGPT input is limited. Please, consider the amount of metadata you will send.
The last use case is to teach ChatGPT a request body for a simple API. We will use its knowledge to fill the request body in fact categorization. Thanks to that we are able to call a method that gives us request data.
