LLM topic and node description examples

Strong topic and node descriptions for large language model (LLM) topics result in a better Virtual Agent user experience.

Topic description examples

Admins write LLM topic descriptions during topic creation and can manage the descriptions on the topic's Properties tab. Topic authors with the virtual_agent_admin role or sn_vad_genai.topic_migration_admin role can also change the LLM-generated topic description on the Publish (optional) page of the topic migration workflow if they want to provide more details than what was automatically generated. In the Topic description field, describe in detail what you want the topic to do. The more information that you provide in your topic descriptions, the more successful your user's experience is with the Virtual Agent. The following table provides examples of both weak and strong topic descriptions.

Table 1. Weak and strong topic description examples

Topic name	Weak topic description	Strong topic description
Company Holiday Calendar	This topic is about a holiday calendar for employees in a company.	This topic is about a holiday calendar for employees in a company. Users can ask for the holiday list or company holiday for a specific year, specific date, inquire about a specific holiday, or ask if they have a day off for a particular holiday. The topic also covers the availability of a holiday calendar and specific holidays like Freedom Day and wellbeing days.
Active Directory Password Change	This topic is related to the management and administration of Active Directory, a directory service provided by Microsoft.	This topic is related to the management and administration of Active Directory, a directory service provided by Microsoft. This topic involves making changes to the Active Directory, such as creating, modifying, or deleting directories and processes. This topic also includes tasks performed by administrators, such as managing user accounts, groups, and permissions within the Active Directory environment.

Node name and Detail description example

Make input node names unique and descriptive so that they explain what operation the user is trying to perform. Descriptive node names enable the LLM to work properly with entities and slot filling. For example, if admins create two similar input node names, such as AD Operation Type and AD Operation, the LLM may identify the wrong node and map it to random input node variables. Renaming the nodes to differentiate the two is best. For example, AD Operation for Password Reset and AD Operation for Locked Credentials and Change Password are stronger, more unique node names.

The Detail description field for all LLM-enabled user input nodes requires you to describe one specific detail that you need from the user. Using detailed descriptions helps the LLM generate user prompts and capture the most important part of their reply.

There are currently seven LLM-enabled user input nodes:

• Input Collector
• Text
• Static Choice
• Dynamic Choice
• Boolean
• Date Time
• Carousel

When working with one of the preceding input nodes, include clear instructions within the Detail description field about what possible values the LLM should request from the user. For example, in an Active Directory Operation topic, admins must be specific in the Detail description field about what operations are allowed. Providing detailed descriptions enables the LLM to use value mapping to identify the entity and fill the slots. The following table provides examples of a weak and strong node Detail description field.

Table 2. Detail description for an Active Directory Operation Type topic's input node example

Weak detail description	Strong detail description
Collect AD Operation Type	Collect Active Directory (AD) Operation Type Additional instructions: For this input these values are mapped values: {"reset":["reset","locked","unable to sign in","expiry", "expired", "expire"],"change":["change"]}. The keys must be extracted if the user input contains the values associated with the keys.

Figure 1. Strong detail description example of LLM-enabled input node