Hi @Sam10
I can help. The short answer is that the LLM is executing "disambiguation" in the synthesized response - asking clarifying questions in order to surface the right content to you.
It has not yet determined a source to answer your question, whether that is a KB Article, a catalog item, or an automation topic that you created in the VA Designer.
You're not directly asking to connect to a human, but I can see that's the intent - you may want to ensure that your Live Agent topic is published. We do have sensitivity filters, but it seems that's not yet configured for you - you may want to set that to the side for now.
You do not have to create a topic in the VA Designer in order to work with the Virtual Agent. If that is what you did, It seems as if you are building a topic that does what the Virtual Agent is designed to do. I'm guessing that based on your reference of the "LLM Topic Node".
Since you're new to the Virtual Agent, let me give you some context:
The VA conversation structure is like this - you have a "setup topic" (Greeting and closing), then you could have LLM VA topics for true automation (reset my password, connect to Live Agent), and you also have content(Knowledge, Catalog). All of that is "discovered" with a skills discovery function, powered by AI Search.
Let's say your user asks something outside your delivered content, the Assistant can direct the user to "Fallback" - create a ticket, or end the conversation. There might be content in your KB that addresses your laptop issue, so the VA is trying. Let's say you ask "When was the Declaration of Independence signed?" that should result in 0 results and your fallback option (you can configure web search to answer that question but that's another topic to set to the side for now).
The key to the new LLM VA experience is the use of AI Search for everything this "Assistant" can do.
This search experience is controlled by the Search Profile in the Assistant configuration (Information Sources). If AI Search is new to you, there is training on SNU and a few academy recordings on AI Search. The search profile has to be published.
Back to your example:
Perhaps the VA has retrieved some laptop troubleshooting information, however, because you're not specific enough, the VA keeps trying to get specifics.
Perhaps respond with some details in order to see a different response.
To see the above in action:
If you have created a VA topic, I would deactivate that topic so the LLM can't find it.
Verify your VA search profile has content (KB and Catalog) that is searchable.
Once you've verified your VA search profile is published, I would pick a KB article and see if the VA can find the knowledge article when you ask a question that you would use to answer your own question. (Sometimes I just ask the title of the KB article)
If it finds the KB article, ask a follow up question or 2 - you can ask open ended questions.
Then I would pick a catalog item that is available from your search profile. (sometimes I just start with "Request a (catalog item name)"
To see if your Live Agent topic is published, and available to your assistant, type in "I want to speak to a human" or "live agent"- that's a delivered topic that is usually available.
You can also test the search results in AI Search Preview.
Hopefully that clarifies how topics work within the LLM VA. "PW Reset" might be the only VA topic (and it's truly for automation) - all other conversational back and forth is because of AI Search.
Hope that helps!
Tricia
Hi @Sam10
I can help. The short answer is that the LLM is executing "disambiguation" in the synthesized response - asking clarifying questions in order to surface the right content to you.
It has not yet determined a source to answer your question, whether that is a KB Article, a catalog item, or an automation topic that you created in the VA Designer.
You're not directly asking to connect to a human, but I can see that's the intent - you may want to ensure that your Live Agent topic is published. We do have sensitivity filters, but it seems that's not yet configured for you - you may want to set that to the side for now.
You do not have to create a topic in the VA Designer in order to work with the Virtual Agent. If that is what you did, It seems as if you are building a topic that does what the Virtual Agent is designed to do. I'm guessing that based on your reference of the "LLM Topic Node".
Since you're new to the Virtual Agent, let me give you some context:
The VA conversation structure is like this - you have a "setup topic" (Greeting and closing), then you could have LLM VA topics for true automation (reset my password, connect to Live Agent), and you also have content(Knowledge, Catalog). All of that is "discovered" with a skills discovery function, powered by AI Search.
Let's say your user asks something outside your delivered content, the Assistant can direct the user to "Fallback" - create a ticket, or end the conversation. There might be content in your KB that addresses your laptop issue, so the VA is trying. Let's say you ask "When was the Declaration of Independence signed?" that should result in 0 results and your fallback option (you can configure web search to answer that question but that's another topic to set to the side for now).
The key to the new LLM VA experience is the use of AI Search for everything this "Assistant" can do.
This search experience is controlled by the Search Profile in the Assistant configuration (Information Sources). If AI Search is new to you, there is training on SNU and a few academy recordings on AI Search. The search profile has to be published.
Back to your example:
Perhaps the VA has retrieved some laptop troubleshooting information, however, because you're not specific enough, the VA keeps trying to get specifics.
Perhaps respond with some details in order to see a different response.
To see the above in action:
If you have created a VA topic, I would deactivate that topic so the LLM can't find it.
Verify your VA search profile has content (KB and Catalog) that is searchable.
Once you've verified your VA search profile is published, I would pick a KB article and see if the VA can find the knowledge article when you ask a question that you would use to answer your own question. (Sometimes I just ask the title of the KB article)
If it finds the KB article, ask a follow up question or 2 - you can ask open ended questions.
Then I would pick a catalog item that is available from your search profile. (sometimes I just start with "Request a (catalog item name)"
To see if your Live Agent topic is published, and available to your assistant, type in "I want to speak to a human" or "live agent"- that's a delivered topic that is usually available.
You can also test the search results in AI Search Preview.
Hopefully that clarifies how topics work within the LLM VA. "PW Reset" might be the only VA topic (and it's truly for automation) - all other conversational back and forth is because of AI Search.
Hope that helps!
Tricia
