How to create and train a chatbot using congluence documentation

[Rodney Espinoza]

Hello Atlassians :-), We have solid confluence documentation but sometimes it is pretty dense I want to create and train a chatbot with the documentation that we have is this possible or do you have guys another workaround?

[Koloman Pfeffer]

Hello @Rodney Espinoza, 

Easiest way to do that is probably to use Atlassian AI, which is included in Confluence Premium and Enterprise. That will of course raise costs for your Confluence instance, but from my experience it works really well.

Additionally, you won't need to manually train a model with the content, as it does it by itself.

But overall it is probably a question of what you want to do with the chatbot afterwards/who should use it. Would it be for internal use or for external/customers?

Edit: And it just came to my mind, that you might want to be careful to feed an AI with your companies data. This normally has to be cleared by your legal team with special focus on data residency in some countries, so not totally straight forward sadly.

[Simon Sahli]

Hi @Rodney Espinoza
we are currently training a Chatbot and setting up an IT self-service portal at the moment, so I can share a bit what I have learned so far.
First, you have to have JSM to make that work. We have Premium, I am not 100% sure what is available on lower subscriptions, but it might also work for them.

Two things, the easy one first. You can setup, in your JSM portal in the project settings in the "Knowledge base" menu, one or multiple Confluence spaces that are considered the knowledge base for your JSM project. 
If someone of your users in the JSM portal begins to type in for example "Printer" in the general search field or also in summary field of the selected request types, then they will find the top 3 knowledge base articles and can open them directly from the portal.

The more work intense, but also more sophisticated approach, is to train the new "Virtual service agent" with intents. You will find a menu for this in the JSM project settings, where you can create intents (common support cases), for which you can define relevant phrases that users are usually looking for regarding these topics. 
After that you can begin to define an individual flow with different ways to solve the intent. Usually you will try to figure out what the problem exactly is and then link to Confluence pages (or better a specific paragraph) where the users can find the solution. 
At the end you will ask if they could resolve the problem and if not then you will want to "escalate" the problem, which means automatically creating an issue based on the provided information of the user.

The configuration of the intents is quite some work, but it might be worth it, if as you say, you have well written Confluence pages that can provide self help to your users.


Give it a try, I like it so far. Let me know if you have any further questions!

Cheers,
Simon

[Kristian Klima]

Hello @Rodney Espinoza and welcome to the Community.

Not sure which chatbot you have in mind but it is possible to integrate OpenAI API with Confluence API and deploy in your own solution.

Our in-product AI assistant is using our Confluence content. It's not a conversational bot, though, as AI tends to get confused the more specific you're trying to be :) 

Or, there are Confluence apps that create a Documentation Center from Confluence content and that you can configure to use your OpenAI API key to generate answers - again, not conversational, but really helpful.