Intent Detection

Intent Detection with the AI Agent

The Agent Assist solution assists agents during live interactions by monitoring conversations and triggering relevant guidance. One of the most effective ways to do this is through intent detection, powered by the Sestek AI Agent's NLU capabilities. By classifying user messages into predefined intents (e.g., RequestRefund or CancelSubscription), the system can understand the true context of a conversation. Unlike keyword matching or rule-based triggers, this approach delivers more accurate and relevant guidance aligned with customer needs.

Step-by-Step Configuration

1. Enter AI Agent Integration Settings in CA Tenant

To enable intent detection in Agent Assist, you must first integrate the AI Agent system with the CA tenant. This is done by filling in the required fields under Administration > Settings > Agent Assist > AI Agent Integration tab.
API address, Identity base URL, Client ID, and secret information are needed for the VA's intent detection feature.

2. Configure AI Agent Project

Create an AI Agent project specifically for the intent detection feature, or use the existing RTG-Integration project flow designed for this purpose. In the existing project, a Condition node is used to filter intent-agent and intent-customer values from the CustomAction field under the Channel Data in the activity.

From the AI Agent tab in the interface, enter the intent and utterances to be used for scenario triggering in Agent Assist.

Intent List:

Utterance List:

An action flow should be designed for each intent to define the appropriate response or guidance, ensuring that the necessary actions are presented to the agents.
For example, in the below action flow for "campaign" type intent, a "say" node is used for action to be presented to the agent such as "offer points" which is derived from the "text" field in the activity object.

3. Create a New Scenario for the AI Agent

Once the AI Agent project is configured, go to the Scenarios section in Agent Assist and create a new scenario. In the Add Trigger section, select AI Agent as the trigger type and enter the name of the relevant VA project.

Once the configuration is complete, the intent detection feature is ready to be used for scenario triggering in Agent Assist.

Work & Data Flow

Example request for AI Agent Synch API Call:

curl --location --request POST 'https://eu.va.knovvu.com/magpie/ext-api/messages/synchronized' \
--header 'Content-Type: application/json' \
--header 'Project: RTG-Integration' \
--header 'X-Knovvu-Conversation-Id: 20250613143138' \
--data-raw '{
  "text": "kampanya",
  "serviceUrl": "http://agent-web.SESTEK-ca/copilot/AgentService/api/v1/Notification/CallBack/20250613143138",
  "conversation": {
    "id": "20250613143138"
  },
  "channelId": "RTG",
  "type": "message",
  "channelData": {
    "CustomAction": "intent-customer",
    "CustomActionData": "{ 
      \"Transcription\": [
        {
          \"ID\": \"439295\",
          \"Sentence\": \"Merhaba, ben Knovvu Sanal Asistan. Size nasıl yardımcı olabilirim?\",
          \"Channel\": 1,
          \"TranscriptTime\": null,
          \"TranscriptDuration\": null
        },
        {
          \"ID\": \"686481\",
          \"Sentence\": \"selam\",
          \"Channel\": 2,
          \"TranscriptTime\": null,
          \"TranscriptDuration\": null
        },
        {
          \"ID\": \"583425\",
          \"Sentence\": \"Canlı temsilci ile iletişime geçiliyor\",
          \"Channel\": 1,
          \"TranscriptTime\": null,
          \"TranscriptDuration\": null
        },
        {
          \"ID\": \"928223\",
          \"Sentence\": \"kampanya\",
          \"Channel\": 2,
          \"TranscriptTime\": \"2024-04-24T11:11:27.169Z\",
          \"TranscriptDuration\": \"0-57:0-22\"
        }
      ]
    }",
    "IsVoiceBased": false
  }
}'

Example response from AI Agent Synch API Call:

{
  "type": "message",
  "id": "0b8c5c4d-5fd5-4190-9a29-29f91707a537",
  "timestamp": null,
  "localTimestamp": null,
  "localTimezone": null,
  "serviceUrl": "http://localhost",
  "channelId": "webchat-external",
  "from": null,
  "conversation": {
    "isGroup": null,
    "conversationType": null,
    "id": "20250612133431",
    "name": null,
    "aadObjectId": null,
    "role": null,
    "tenantId": null
  },
  "text": "Puan teklif et!",
  "speak": null,
  "inputHint": null,
  "summary": null,
  "suggestedActions": null,
  "attachments": [],
  "entities": [],
  "channelData": {
    "IsSynchronizer": true,
    "HasExternalAudioContent": false,
    "SrConfidence": null,
    "CustomProperties": {
      "Checklist": "NULL",
      "Intent": "Campaign",
      "Confidence": "0.995",
      "ActionType": "1"
    },
    "CustomAction": null,
    "CustomActionData": null,
    "ProjectName": "RTG-Integration",
    "ResponseType": 0,
    "EndUser": {
      "Name": null,
      "Phone": null,
      "Email": null,
      "Twitter": null,
      "Id": null
    }
  },
}