Some webhook calls may fail and fall back to static response

[ttzeng]

The webhook response in Diaglogflow must within 5 seconds, or the request will time out. Right now the input.unknown action relays the requests to OpenAI API, that may fail and the static responses are return.

[ttzeng]

A typical example, a row interaction log from the Dialogflow ES console:

{
  "id": "8d9aebfb-bc0a-4090-ac21-ecb32e30607b-fb923cf8",
  "fulfillmentText": "對不起，我聽不懂你的問題。",
  "language_code": "zh-tw",
  "queryText": "如何投資海外債券",
  "webhookPayload": {},
  "intentDetectionConfidence": 1,
  "action": "input.unknown",
  "webhookSource": "",
  "parameters": {},
  "fulfillmentMessages": [
    {
      "text": {
        "text": [
          "對不起，我聽不懂你的問題。"
        ]
      }
    }
  ],
  "diagnosticInfo": {
    "webhook_latency_ms": "4893.0"
  },
  "webhookStatus": {
    "webhookStatus": {
      "code": 4,
      "message": "Webhook call failed. Error: DEADLINE_EXCEEDED, State: URL_TIMEOUT, Reason: TIMEOUT_WEB."
    },
    "webhookUsed": true
  },
  "outputContexts": [
    {
      "lifespanCount": 1,
      "name": "__system_counters__",
      "parameters": {
        "no-input": "0.0",
        "no-match": "8.0"
      }
    }
  ],
  "intent": {
    "isFallback": true,
    "displayName": "Default Fallback Intent",
    "id": "53efae5c-f5b1-402c-be69-34f52c5d58c7"
  }
}

[ttzeng]

The OpenAI may take time to respond longer context, TIMEOUT still occurs if the overall elapsed time exceeds the fulfillment requirement:

Webhook input from Dialogflow:

{'responseId': '9ee05192-88bb-498f-89a9-4e244cb4fc4f-fb923cf8', 'queryResult': {'queryText': 'No, I meant a template of LINE message API using openai', 'action': 'input.unknown', 'parameters': {}, 'allRequiredParamsPresent': True, 'fulfillmentText': '對不起，我聽不懂你的問題。', 'fulfillmentMessages': [{'text': {'text': ['對不起，我聽不懂你的問題。']}}], 'outputContexts': [{'name': 'projects/ttzeng-gcp/agent/sessions/801e75c5-6920-3c28-8c0b-d933e45206cf/contexts/__system_counters__', 'lifespanCount': 1, 'parameters': {'no-input': 0.0, 'no-match': 3.0}}], 'intent': {'name': 'projects/ttzeng-gcp/agent/intents/53efae5c-f5b1-402c-be69-34f52c5d58c7', 'displayName': 'Default Fallback Intent', 'isFallback': True}, 'intentDetectionConfidence': 1.0, 'languageCode': 'zh-tw'}, 'originalDetectIntentRequest': {'source': 'line', 'payload': {'data': {'replyToken': '84ec21db19f6457989f68c514080b87c', 'type': 'message', 'source': {'type': 'user', 'userId': 'U83c6e8f0e536cbca2a61f0eb8a1f6951'}, 'timestamp': '1717990700058', 'message': {'type': 'text', 'id': '511972510029054098', 'text': 'No, I meant a template of LINE message API using openai'}}}}, 'session': 'projects/ttzeng-gcp/agent/sessions/801e75c5-6920-3c28-8c0b-d933e45206cf'}

Response from OpenAI in the webhook:

Here is a template for using the LINE messaging API with OpenAI:
```python
import requests
# Set up LINE messaging API credentials
LINE_API_URL = "https://api.line.me/v2/bot/message/push"
LINE_ACCESS_TOKEN = "YOUR_LINE_ACCESS_TOKEN"
# Set up OpenAI API credentials
OPENAI_API_URL = "https://api.openai.com/v1/engines/davinci/completions"
OPENAI_API_KEY = "YOUR_OPENAI_API_KEY"
# Define function to send message using LINE API
def send_line_message(user_id, message):
    headers = {
        "Authorization": f"Bearer {LINE_ACCESS_TOKEN}",
        "Content-Type": "application/json"
    }
    data = {
        "to": user_id,
        "messages": [
            {
                "type": "text",
                "text": message
            }
        ]
    }
    response = requests.post(LINE_API_URL, headers=headers, json=data)
    return response.json()
# Define function to generate response using OpenAI API
def generate_openai_response(prompt):
    headers = {
        "Authorization": f"Bearer {OPENAI_API_KEY}",
        "Content-Type": "application/json"
    }
    data = {
        "prompt": prompt,
        "max_tokens": 100
    }
    response = requests.post(OPENAI_API_URL, headers=headers, json=data)
    return response.json()["choices"][0]["text"]
# Example usage
user_id = "USER_ID"
prompt = "Hello, how can I help you today?"
response = generate_openai_response(prompt)
send_line_message(user_id, response)

In this template, you can replace YOUR_LINE_ACCESS_TOKEN and YOUR_OPENAI_API_KEY with your actual credentials. You can then use the send_line_message function to send a message to a specific user on LINE, and the generate_openai_response function to generate a response using OpenAI's language model.

Row interaction log in the Dialogflow ES console:

{
  "id": "9ee05192-88bb-498f-89a9-4e244cb4fc4f-fb923cf8",
  "fulfillmentText": "對不起，我聽不懂你的問題。",
  "language_code": "zh-tw",
  "queryText": "No, I meant a template of LINE message API using openai",
  "webhookPayload": {},
  "intentDetectionConfidence": 1,
  "action": "input.unknown",
  "webhookSource": "",
  "parameters": {},
  "fulfillmentMessages": [
    {
      "text": {
        "text": [
          "對不起，我聽不懂你的問題。"
        ]
      }
    }
  ],
  "diagnosticInfo": {
    "webhook_latency_ms": "4893.0"
  },
  "webhookStatus": {
    "webhookStatus": {
      "code": 4,
      "message": "Webhook call failed. Error: DEADLINE_EXCEEDED, State: URL_TIMEOUT, Reason: TIMEOUT_WEB."
    },
    "webhookUsed": true
  },
  "outputContexts": [
    {
      "lifespanCount": 1,
      "name": "__system_counters__",
      "parameters": {
        "no-input": "0.0",
        "no-match": "3.0"
      }
    }
  ],
  "intent": {
    "isFallback": true,
    "displayName": "Default Fallback Intent",
    "id": "53efae5c-f5b1-402c-be69-34f52c5d58c7"
  }
}