Hi all, I am hoping to find a quick solution for this. This assistant's task is to collect customer information for functions. She has 8 pre-qualified questions/info to collect. When there is some kind of interruption (e.g background noise) or a subject change (e.g custom asks a different question), she does not go back to complete the pre-qualified questions (information is not asked and collected) and she does not trigger to send email. For this specific call below, we also have result as "Success" but it was not as it did not complete the task: { "name": "endCall", "role": "tool_call_result", "time": 1740112040461, "result": "Success.", "toolCallId": "call_obuZtghCpRIHPLq6hicfKQzn", "secondsFromStart": 198.21 } CALL ID: feff38ad-9a15-4a35-96b6-a60f39bde87d Timestamp: 02:58 What we expect to happen: Assistant goes back and continues the Pre-Qualified questions, ends the call and fires off email trigger What happened: Assistant "forgets" where she was at, starts a whole new conversation "Hello, thanks for calling, how can I help you?", says she called at the wrong time (when she didnt make the call as it was inbound), ends the call abruptly and does not fires off email trigger.

The issue stems from two main areas: \- Context Management: The assistant isn't maintaining state of collected information \- Interruption Handling: Despite having interruption handling instructions, the assistant failed to return to the previous context You need to write your assistant prompt: 1\. Maintain state throughout the conversation 2\. Handle interruptions gracefully 3\. Ensure complete information collection

examples

Enhance the prompt to include context persistence:

const CONTEXT_MANAGEMENT = `
[Context Management]
- Maintain a running list of collected information
- On interruption, store current progress
- After handling interruption, explicitly return to last incomplete item
- Before ending call, verify all required information is collected
- Only trigger endCall after either:
  a) All information is collected and functionsEnquiry is called
  b) Customer explicitly ends conversation
  c) Error condition requires termination
`;

Add state tracking to the assistant's prompt:

const STATE_TRACKING = `
[Information Tracking]
Required Fields:
1. numberOfGuests: [STATUS]
2. firstName: [STATUS]
3. lastName: [STATUS]
4. email: [STATUS]
5. phoneNumber: [STATUS]
6. functionDate: [STATUS]
7. catering: [STATUS]
8. budget: [STATUS]
9. specialRequests: [STATUS]

On interruption:
1. Save current state
2. Handle interruption
3. Return to last incomplete field
4. Confirm remaining information before ending call
`;

\`\`\` it's more of prompting task with different approaches trial and error but I will try to implement few by myself so adding your ticket in my investigation list.

Thank you, will try this!

@firefly_0206 could you let me know if the prompt was able to help? We are running into similar issues. 1. The Assistant does not maintain the context and switches to a new subject. 2. Experiencing delays of 5 seconds or more from the AI assistant response. 3. Assistant glitches and cuts out of the conversation. It sounds almost like it is losing signal. 4. It does not handle interruptions gracefully. When the user is trying to speak the Assistant does not stop. Call ID for support: 9dc7071b-b2b5-4bc5-8a92-86fd69722c42 Timestamps of issues: 0:55 - ai doesn’t respond to question 1:58 - cuts itself off in the thought 2:08 - doesn’t stop talking when interrupted by user 3:50 - long pause after user finishes before ai speaks 4:04 - so assistant had 2 thoughts 4:21 to 5 - ai assistant interrupting itself a lot 5:40 - glitching like cell phone signal loss 6:09 - glitching like cell phone signal loss 6:30 - glitching like cell phone signal loss 7:00 - glitching like cell phone signal loss 7:24 - ai assistant cuts off 7:46 - ai assistant glitches and cuts off

Will put an update in here once we apply and test it. Havent yet gotten to this!

@Tim Tran can you please create a #1211483291191083018 ticket and ping me over there.

@Tim Tran Just an update - we've used the tracking prompt suggest and see a big improvement on coming back to collecting info after interruption