Using AI Knowledge Sets

Overview

When building out Conversational AI virtual service agents (VSAs) and AI Assist Agents, Knowledge Sets allow both products to use Generative AI in the context of your organization's information.

Where to Find This

This feature appears in two locations, depending on your use case:

  • Administration > Knowledge Sets
    • Knowledge sets that are global to the system and used by both Conversational AI and Assist Agents
  • In an AI Assist Agent Application > Knowledge Sets
    • Knowledge sets that are scoped to one specific Assist Agent

Once navigated to, a table of currently available knowledge sets is shown to the user. From this view, users may create new, delete, or manage existing knowledge sets.

AI Knowledge Sets Capabilities

Once an AI Knowledge Set has been created or existing one clicked on, users are presented with 4 possible screens, each presenting their own unique purpose: Configuration,  Sources,  Terms, Activity, & Metrics. Below is an outline of the available screens & their respective functions.

Configuration

The configuration screen is the primary control panel for where AI Knowledge Sets can be manipulated at large. On this screen users will be able to define many variables such as subject matter, description, indexing options, custom role restriction and scheduled updates.

  • Subject Matter
    • A brief description of the subject matter or context of the content of this knowledge set. Providing this is optional but can result in better knowledge query results for general or vague questions.
    • Just a few words works best. For example, if the knowledge set included a set of articles about quantum physics, you would use "quantum physics".
    • This essentially defines the subject matter with which this knowledge set will be an expert on.
  • Indexing Options
    • This is where adding new content, deleting content, updating content, removing content & viewing information about the knowledge set can occur.
    • Clear: Clears the index within the underlying Knowledge Store. All content will be removed.
    • Ingest: Selected knowledge sources can be ingested as net-new content made available to the AI Knowledge Set here.
    • Refresh: Selected knowledge sources can be deleted entirely, then re-ingested for a full refresh when needed.
    • Update: Selected knowledge sources can have a manual update ran against them, based on the update policies set at the source level.
    • Remove: Selected knowledge sources are removed from the knowledge set.
    • Index Info: Provides basic information about the information that has been indexed (or ingested) into the current knowledge set.
  • Scheduled Updates
    • Allows administrators to set update schedule policies for various knowledge sources contained within the knowledge set. This can be a one time or recurring scheduled update. 
  • Scheduled Refreshes
    • Allows administrators to set refresh schedule policies for various knowledge sources contained within the knowledge set. This can be a one time or recurring scheduled refresh. 

Sources

The sources screen is the area with which builders will define the sources that will be ingested. Available sources include web crawl, TeamDynamix KB articles, TeamDynamix Tickets, FTP files, and local files. Sources are effectively the various art facts of content that will be housed within the knowledge set.

  • Web Crawl
    • Users may define one to many URLs to be crawled to be ingested into the knowledge set. Maximum requests, crawl strategy, keywords meta tag, and update policy may be defined before clicking the 'advanced configuration' button.
    • Within the advanced configuration screen builders may configure URL filters, include selectors, exclude selectors, basic authentication login information, and indexing options.
    • Users may also add login method(s) to their web crawl - Login Page & Script
      • Login Page: Allows users to specify a login page URL, username, password, and login button CSS selector. Conversational AI will then script a login procedure to access authenticated web content.
      • Script: Allows users to build a series of scripted steps without writing code. Users may script one to many actions of the following type to complete a login process to allow Conversational AI access to authenticated web content.
        • Navigate: will navigate the crawler to a specified URL, typically a login page.
        • Fill: Allows fields on a page to be filled via script. Useful for fields like username and password.
        • Click: Allows buttons on the navigable page to be clicked by the script.
        • Wait: Will create a wait condition between other script actions.
  • TeamDynamix KB Articles & Tickets
    • Users will select an existing (or newly created) set of shared credentials that have access to TDX article reports in TDNext. Two articles will be required to configure this type of source: an initial ingestion KB article report, and an update article report. This approach was taken to ensure that the flexibility of the work management reporting system isn't missed or rebuilt entirely within CAI, allowing folks maximum flexibility and control over content that will be ingested.
    • Initial ingestion article report:
      • The report to use to fetch either the Ticket or Article Ids, depending upon the selected source type, that should be ingested into the knowledge set.
      • For tickets, the report should contain the following fields: TicketID and Application. For articles, the report should contain the following fields: ID and Primary Client Portal
      • Tickets also have the option to toggle to include or exclude the feed during ingestion. This can be useful for identifying resolution details that may have been provided in an update or comment.
    • Update article report:
      • The report to use to fetch either the Ticket or Article Ids, depending upon the selected source type, when updating this source, either manually or due to a scheduled update.
      • The report must include, at a minimum, the ticket/article id and the application name.
      • Typically, this report would be similar to the regular report used for initial ingestion except that it would filter the results to tickets/articles that have been updated recently.
  • SFTP Files
    • Users will select credentials, host & path to point to files desired to be pulled from via FTP. An update since rule may also be configured here. For row-based files, an additional section "record-based file information" is also present to skip headers, make specific references, and more.
  • Local Files
    • Users will select file(s) from their local machine to be ingested into the knowledge set directly. Update policies may not be set for this source type since there will be no direct connection created between CAI and the machine (for these capabilities, please use the FTP Files source type).
  • Knowledge Set Flow
    • Requires an iPaaS license. The system will run the selected Knowledge Ingestion Flow, which is a special kind of flow that returns a set of text records to index. You can use this type of source to include almost any source of text content in a knowledge source.
  • iPaaS Flow Using Knowledge Set Connector
    • Requires an iPaaS license. Similar to an iPaaS Flow source, a Direct source is populate via an iPaaS flow. You can add content to a direct source from any flow using the Knowledge Set Connector.
  • TD Work Management Response Templates
    • The system will ingest specified TeamDynamix Work Management ticket response templates for the specified work management application.
    • Provide credentials to connect to the work management instance, include the application ID to pull response templates from, and our job scheduler will handle the rest.

Terms

The terms screen is the area in which builders will define terms or words that would be desirable to have replaced before passing the user's utterance to the LLM for a generative answer. The benefit of this capability is ensuring that the 'persona' of the VSA is well-defined so that ambiguous questions that include words like "you, your, our, etc." can be accounted for and replaced with the appropriate persona.

Activity

The Activity screen is the area in which builders will view the progress of ingestion jobs of any type. It shows what is being added, removed, or modified within the knowledge set. This screen gives information on the name of the job, type of job, created and completed time, applicable messages, status, and ID.

From this screen using the dropdown arrow next to a job name, the log of events may be viewed or downloaded. Users may also cancel an in-progress job from this screen.

Metrics

The Metrics screen is the area in which users may view usage data and information specific to the currently opened knowledge set. Think of this as a window into usage (successful and failed) for your knowledge sets. There are many charts present on the Metrics screen, of which are listed below:

  • Knowledge Set Usage (successes & failures): Line graph depicting usage over the selected time range.
    • All Time, Today, Yesterday, Last 7 Days, Last Month, This Month, This Year & Last Year
  • Daily Knowledge Set Usage (successes & failures): Bar charts depicting daily usage over the selected time range.
    • Last 7 Days & Last Month
  • Monthly Knowledge Set Usage (successes & failures): Bar charts depicting monthly usage over the selected time range.
    • This Year & Last Year
Print Article

Related Articles (2)

Creating and Configuring Knowledge Sets
This article reviews Knowledge Sets, and how to create them