It works pretty nice but I am trying to find out if you know if it only respond to question on my knowledge base or it also answers on his own?
do you have any experiment with it?
https://marketplace.atlassian.com/apps/1230358/kami?tab=overview&hosting=cloud
Thanks..
Welcome to the Atlassian Community!
The AI chat bots I've tried out are superbly awful. At the time I type this, TLDR: "No"
But.
They can work. I've had some good answers from AI 'bots.
I think the main problem is what they have been trained on. ChatGPT and OpenAI work ok for general questions that you might type into google, bing, duckduckgo and so on (and frankly, their search teams need to be scared of this), but they are utterly s***e at specialised subjects.
If you can get a chat AI trained on useful information, then yes, use them. But most are not (yet)
It's pretty handy when it comes to optimizing SQL queries but it needs some Jira training!
Hi,
My name is Mushon I am part of the team that built that plugin. actually we are using OpenAI Davinci model to less get disrupt of chat api. What we did is chunking the most relevant data out of Confluence (using embedding) that best matches the request and doing some sophisticated prompting to push OpenAI model to answer only based the text provided.
I would love to hear what you think?
Hey @Polin Green you are invited to reach me if you would like to get mode details: mushon@okami.io
Thanks :)
I've built a Chatbot Integration using Landbot and webhooks to ChatGPT and Confluence that works using our Knoweldgebase space as the ChatGPT data source added to a prompt.
I built one prompt for a webhook that extracts the keywords from the user's phrase and i built a second prompt to summarise a Confluence Page after displaying article page titles that match on a title~ [keyword*] search using the keywords, found from the user's search phrase.
So my bot lets a user type in something like "how do i print off a list of applicants", as a search phrase. ChatGPT breaks that down to be the keywords, print+list+applicants. and ChatGPT also delimits the keywords needed for the querystring to GET the pages from Confluence REST API. I also strip out all the unwanted garbage from the user search phrase leaving pure keywords.
The user then picks the nearest matching article title from a list I output as text buttons with the titles headlineing each button in my chatbot, and then ChatGPT summaries the page in the article lookup, into a simple list of steps 1,2,3 for the user to follow rather than them having to read the whole article.
It works pretty well for me at the moment so I am going to release it as a BETA to my public user base.