When a customer dials your agent's phone number, the call flows through a fully automated pipeline -- from the telephony network through the AI voice processing engine and back. The entire interaction is recorded, transcribed, and analyzed automatically.
Here is the step-by-step journey of an inbound call:
Caller dials +91-XXXXXXXXXX
|
v
Telephony network (Indian telecom partner)
|
v
Voice AI engine (WebSocket connection)
|-- Speech recognition transcribes caller speech to text
|-- AI language model generates response using system prompt + knowledge base
|-- AI voice synthesis converts response text to natural speech
|
v
Caller hears AI response
|
(conversation continues until resolution or escalation)1. Connection and Greeting
When the call connects, the agent immediately plays its greeting message. This is the greetingMessage you configured in Step 2 (Configure) of the setup wizard. A typical greeting:
"Hello, thank you for calling Acme Corp. This is Maya, your AI assistant. How can I help you today?"
2. Conversation
The caller and agent converse naturally. The agent:
This loop continues for each exchange. The agent maintains conversation context throughout the call, remembering what was discussed earlier.
3. Knowledge Base Lookup
When a caller asks a question, the agent automatically searches your knowledge base for relevant content. The top matching chunks (up to 5) are injected into the LLM context, allowing the agent to answer with your specific business information rather than generic responses.
4. Resolution or Escalation
The call ends in one of three ways:
Every call is automatically:
Recordings and transcripts are valuable for quality assurance, training refinement, and dispute resolution.
After each call, the system analyzes the conversation to determine caller sentiment. This helps you identify:
Sentiment data appears in your call logs and contributes to the analytics dashboard.
All calls are logged in the call_logs table with the following information:
| Field | Description |
|---|---|
| Call ID | Unique identifier from the voice platform |
| Duration | Length of the call in seconds |
| Transcript | Full conversation text |
| Sentiment | Positive, neutral, or negative |
| Summary | AI-generated summary of the call |
| Caller Number | The phone number that called (when available) |
| Timestamp | When the call started |
| Handoff Status | Whether the call was escalated |
Access call logs from the My Agents page in your customer portal or the Call Logs section in the admin panel.
The three-level escalation system works as follows:
At every level, the agent always asks for consent before transferring. It never silently transfers a caller.
Tip: Monitor your call logs regularly during the first few weeks after going live. Look for calls where the agent could not answer or where sentiment was negative -- these indicate areas where your knowledge base needs more content.
The voice platform sends webhook events to your Zentrix Agent Studio instance at key points during and after calls:
These webhooks are handled automatically by the /api/calls/webhooks endpoint. See Webhooks for details on the event format.
A Product by BRTNeura Technology LLP
Last updated: 2026-03-05