Step 2: Connect OpenAI Requests for Classification and Response

These HTTP requests call OpenAI to classify the question and generate the final answer.

1. In Categorize Inquiry, set URL to https://api.openai.com/v1/chat/completions and keep Send Body and Send Headers enabled.
2. Set Body Parameters for Categorize Inquiry:
   • model = gpt-3.5-turbo
   • messages = =[{"role": "system", "content": "You are a travel question classifier. Classify the following travel question into ONE of these categories: DESTINATIONS, PACKAGES, VISA, TRANSPORT, HOTELS, ACTIVITIES, BOOKING, CANCELLATION, BAGGAGE, GENERAL. Respond with only the category name in uppercase."}, {"role": "user", "content": "{{ $json.userQuestion }}"}]
   • temperature = 0.2
   • max_tokens = 20
3. In Compose AI Response, set URL to https://api.openai.com/v1/chat/completions with Send Body and Send Headers enabled.
4. Set Body Parameters for Compose AI Response:
   • model = gpt-4
   • messages = =[{"role": "system", "content": "You are a helpful and friendly airlines travel assistant. Answer travel questions accurately based on the provided context. Be concise but informative. Use a warm, professional tone. If you don't have specific information, provide general guidance and suggest contacting customer support for details. Always prioritize customer satisfaction and safety.\n\nContext Information:\n{{ $json.knowledgeContext }}"}, {"role": "user", "content": "{{ $json.userQuestion }}"}]
   • temperature = 0.7
   • max_tokens = 500

Step 3: Set Up Question Normalization and Category Context

The code nodes clean input, interpret the classifier output, and inject the right FAQ context.

1. In Normalize User Question, keep the provided JavaScript to extract question, user_id, and session_id from the webhook payload.
2. Connect Normalize User Question to Categorize Inquiry to pass userQuestion and metadata to the classifier.
3. In Read Category Output, keep the JavaScript that reads $input.item.json.choices[0].message.content and references Normalize User Question with $('Normalize User Question').item.json.
4. In Retrieve FAQ Context, keep the knowledge base map and ensure it uses const relevantContext = knowledgeBase[category] || knowledgeBase['GENERAL'];.

Step 4: Assemble the Reply and Route Satisfaction

This step generates the final payload, checks satisfaction keywords, and branches to the appropriate follow-up logging path.

1. In Assemble Reply Payload, keep the JavaScript that builds answer, relatedLinks, and usage fields based on Retrieve FAQ Context.
2. In Assess Satisfaction, set the condition to match the expression ={{ $json.userQuestion.toLowerCase() }} against the regex thank|thanks|helpful|great|perfect|excellent|satisfied.
3. Connect the true output of Assess Satisfaction to Record Happy User and the false output to Provide Human Assistance.
4. Ensure both Record Happy User and Provide Human Assistance connect into Combine Satisfaction Routes with Mode set to combine.

Step 5: Configure Logging and Webhook Response

Log each interaction and return the final response to the original webhook caller.

1. In Persist Interaction Log, set URL to https://your-database-api.com/logs/interactions and keep Send Body and Send Headers enabled.
2. Set the Body Parameters in Persist Interaction Log using expressions:
   • userId = ={{ $json.userId }}
   • sessionId = ={{ $json.sessionId }}
   • question = ={{ $json.userQuestion }}
   • category = ={{ $json.category }}
   • answer = ={{ $json.answer }}
   • satisfaction = ={{ $json.satisfactionStatus }}
   • timestamp = ={{ $json.timestamp }}
   • tokensUsed = ={{ $json.tokensUsed }}
3. In Return Reply to Client, set Respond With to json and keep Response Body as: ={{ { "status": "success", "answer": $json.answer, "category": $json.category, "relatedLinks": $json.relatedLinks, "followUpMessage": $json.followUpMessage || null, "supportOptions": $json.supportOptions || null, "sessionId": $json.sessionId, "timestamp": $json.timestamp } }}

Step 6: Test and Activate Your Workflow

Verify the entire path from webhook intake through AI response and logging, then enable the workflow.

1. Click Execute Workflow and send a test POST request to the Incoming Query Webhook URL with a JSON body containing question and user_id.
2. Confirm the execution flow follows Incoming Query Webhook → Normalize User Question → Categorize Inquiry → Read Category Output → Retrieve FAQ Context → Compose AI Response → Assemble Reply Payload → Assess Satisfaction → Combine Satisfaction Routes → Persist Interaction Log → Return Reply to Client.
3. Verify the response includes answer, category, and relatedLinks, and optionally followUpMessage if routed through Provide Human Assistance.
4. Once confirmed, toggle the workflow to Active to handle live questions.