Agent Assist

Prev Next

Integration node for Agent Assist, also known as Real-Time Guidance (RTG).

Parameters

name
description
default
Address Address of the Conversational Analytics ws://ca-external-api.sestek-ca/
IgnoreSslErrors Ignore any certificate errors if Conversational Analytics address contains https false
CredentialId Id of the credential that will be used for Conversational Analytics http calls. None
Send Offline Transcripts To CA If you want to send transcripts to CA for offline analysis, check this box. For more information, check Project Requirements false
Language Language Code for the conversation, sent to RTG for context. en-US
Model Model information for AI Generated Scenarios.
MaxTokens Maximum amount of tokens to be used by the provider for AI Generated Scenarios. 2048
Temperature Temperature Value to be used by the provider for AI Generated Scenarios 0.0
VLLM API Base VLLM API Base, required if provider is VLLM
Provider Name of the LLM provider for AI Generated Scenarios VLLM
AI Generated Scenarios Credential API Key Credential to be used by the LLM Provider.
Azure API Base API Base string, required if the provider is Azure
Azure API Version API version, required if the provider is Azure
How Many Turns To Send To Generative AI How many turns should be collected before checking for AI Generated Scenarios 4

Inputs

Audio

 One Channel Audio Input

Events

name description known nodes that generate this event
SR Milestone Speech Recognition results, used for wildcard phrase checks in word scenarios, LetterPerSecond generation in speed scenarios, and transcript generation. SR Http, SR WebSocket
Emotion Audio Emotion results. The node checks if monotonicity is outside normal ranges depending on the scenario. Emotion
Sentiment Text Sentiment results. Used in transcript generation on Customer's channel, and negative sentiment scenarios. Sentiment
CallStateUpdate Updates CA Conversation Status depending on Call State, can be Start, End, On Hold, Continue etc. Entry, External Relay
Chat Message Used for text-only calls. Works similarly to SR Milestone. Doesn't trigger speed scenarios. Entry, External Relay
Translation Translation results, can be used in place for SR Milestone and Chat Message if the project is VT. Translate ML, Translate Google, Translate OpenAI

Outputs

Audio

 none

Events

default description
RTG Scenarios This event contains the active and filtered scenarios. The scenarios are contained within an array. This event is generated at startup.
RTG State This event contains the Conversation State updates. These events are generated when the conversation state changes.
RTG Notification This event is generated whenever a notification is triggered, and contains the notification information.
RTG Transcript This event occurs whenever a transcript is generated and has sentiment information if it is a customer sentiment.
Change Insights Updates the Conversation Insights node depending on RTG Settings.
RTG Supervisor Message The messages sent by the supervisor are sent through this node.

Remarks

Requirements

License

RTG License

The agent for the call must be licensed for RTG in CA Interface.

image.png

image.png

To check the active licenses of agents, you can go to the Administration->License Management -> User Licenses tab in CA Web Interface. The RTG Licenses are under the RTG Tab.

"StartMessage" requirements

"StartMessage" should contain

  • At least one definition under "CallProperties.UserProperties" except for AgentName. This lets the node to determine the agent and corresponding scenarios in this session. Since AgentName cannot be used to determine the agent, we need one of the other unique identifiers.
  • Definitions for ChannelTags. This specifies which channel belongs to the customer and which to the agent.
  • Other fields are optional.
  {
    "CallProperties": 
    {
      "ExternalId":"", // string 
      "UserProperties":
      {
        "AgentName":"",     // string
        "DeviceId":"",      //string
        "ExternalId":"",    //string
        "PbxAgentId":""     //string
      },
      "CustomerPhoneNumber": "",    // string 
      "Direction": 1,               // integer 
      "CallerNumber":"",            // string 
      "CalledNumber":"",            // string
      "CallStartTime":""            // ISO 8601 DateTime formatted string
    },
    "CA":{
        "TextBased": false // True if the call is text only.
    }
    "Settings":
    {
      "ChannelTags": 
      [
        "a_tag_for_channel_0",                          //string seperated by "," or ";"
        "a_tag_for_channel_1;another_tag_for_channel_1" //string separated by "," or ";"
      ]
    }    
  }

"StartMessage" example

  {
    "CallProperties": 
    {
     "ExternalId":"a_valid_call_id",
     "UserProperties":
      {
        "PbxAgentId":"a_valid_pbx_agent_id"
        "ExternalId":"a_valid_external_id"
      },
      "CallStartTime":"2022-10-31T09:00:00Z"
    },
    "Settings":
    {
      "ChannelTags": 
      [
        "agent",
        "customer"
      ]    
    }
  }

Workflow

This node has many responsibilities to make sure Agent Assist (RTG) works properly.

  • Pulls RTG Settings and configures Conversation Insights.
  • Pulls and filters the applicable RTG Scenarios.
  • Analyzes each event for Speed, Monotonicity, Sentiment ,Phrase Matches, and Checklists.
  • Sends Transcripts to RTG Dashboard.
  • Sends audio input to RTG Dashboard for Live-Listen when requested.
  • Sends "RTG Supervisor Message" event when a supervisor message is received.
  • Sends transcripts to Generative AI for AI Generated Scenario Analysis every few turns based on the node configuration.

Moments

While information is flowing to RTG, call related events can be observed in the RTG Moments board.

image.png

RTG Dashboard

At the start of session, the node notifies RTG that the determined agent is on a call. Which changes the Conversation Status of the agent in RTG Dashboard.
Upon receiving CallStateUpdate events from the client, or when the call ends, the node informs RTG to change the conversation Status accordingly.

image.png

Transcripts

Agent assist node provides transcripts to RTG Dashboard in real time. It also includes customer side sentiment information in the transcripts if available.

image.png

Project Structure

The Agent Assist node requires a Channel Switch node. Input from two SR Http nodes, Emotion from the Agent channel, and Sentiment from the Customer channel.

For Text-Based calls, a Text Correction node is recommended. The incoming "Chat Message" events will also enter sentiment recognition for the customer channel.

Only one of "SR Milestone", "Translation", and "Chat Message" events should be received per channel. So, if the project uses a Translation Node, the SR nodes shouldn't be directly connected to the Agent Assist node.

Image

The default project ca-online in Data Flow contains a ready to use Agent Assist node.

Supported flow types

Stream

Release Notes

v4.10.0
  • Introduced Node.