CLU endpoint call returns 404 and bot variable clu_result is not defined during Switch step

Hi Suma Nalluri

Welcome to Microsoft Q&A and Thank you for reaching out.

The error means the Language Understanding node never executed successfully, so the variable clu_result was never created. Common causes include:

1. Incorrect variable mapping
   • The output variable name in the LU node must exactly match what you reference later (case-sensitive).
   • If you reference clu_result.result.prediction.topIntent, ensure the LU node output is clu_result.
2. LU node not triggered
   • If the LU node is conditional or disconnected in the scenario flow, it won’t run.
   • Check the scenario graph to confirm the LU node is linked and reachable from the prompt step.
3. Failed API call to CLU endpoint
   • Even if the configuration looks correct, Health Bot will skip variable creation if the CLU call fails.
   • Causes: wrong endpoint URL, invalid key, incorrect project/deployment name, or mismatched API version.
4. Timeout or empty response
   • If CLU returns an empty or malformed response, Health Bot does not create the variable.

1. Validate CLU Node Configuration

• In the LU node:
  • Endpoint: Should be https://<region>.api.cognitive.microsoft.com/language/:analyze-conversations?projectName=<project>&deploymentName=<deployment>&api-version=2022-10-01-preview (or latest supported version).
  • Subscription Key: Matches the Azure Language resource key.
  • Project & Deployment Names: Exactly as in Language Studio.

API Version: Use the version supported by Health Bot (check Azure Health Bot CLU integration guide).

2. Check Variable Mapping

• In the LU node settings:
  • Output variable name must be clu_result (or whatever you reference later).
  • Input text should be mapped to freeText or the variable holding user input.

3. Add Error Handling

If clu_result exists
    → Switch on clu_result.result.prediction.topIntent
Else
    → Fallback message: “Sorry, I couldn’t understand that.”

This prevents runtime crashes if CLU fails

Test CLU Endpoint Independently

curl -X POST "https://<region>.api.cognitive.microsoft.com/language/:analyze-conversations?projectName=<project>&deploymentName=<deployment>&api-version=2022-10-01-preview" \
-H "Ocp-Apim-Subscription-Key: <your-key>" \
-H "Content-Type: application/json" \
-d '{"kind":"Conversation","analysisInput":{"conversationItem":{"id":"1","text":"What is the dosage for aspirin?","modality":"text","language":"en","participantId":"1"}},"parameters":{"projectName":"<project>","deploymentName":"<deployment>","stringIndexType":"

Enable Debug Logging

In Health Bot scenario editor:

• Turn on Trace and check if the LU node shows “Call for LU model failed.”
• If yes, inspect the error details (often authentication or endpoint mismatch)

Reference:

https://learn.microsoft.com/en-us/azure/ai-services/language-service/conversational-language-understanding/how-to/quickstart-multi-turn-conversations?wt.mc_id=knowledgesearch_inproduct_azure-cxp-community-insider#configure-roles,-permissions,-and-settings

https://learn.microsoft.com/en-us/azure/machine-learning/how-to-debug-parallel-run-step?view=azureml-api-1

https://learn.microsoft.com/en-us/azure/machine-learning/how-to-configure-cli?view=azureml-api-2&tabs=public

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Hi Suma Nalluri

Did you get any chance to review the above response. Thank you!

Hi Suma Nalluri

please follow the steps

1. Check the Endpoint and Region

Ensure you are using the correct Language resource endpoint. It should match the region where your CLU project is deployed.

Example:

https://<your-resource-name>.cognitiveservices.azure.com

or the regional vanity URL:

https://language.cognitive.azure.com

Mismatch between endpoint region and project deployment will cause errors.

2. Validate API Version

CLU runtime calls require the correct api-version in the URL:

POST /language/:analyze-conversations?api-version=<version>

Use the latest GA version from Azure documentation. An incorrect or missing version often results in 404 or 400 errors.

3. Confirm Authentication

Use the Language resource subscription key, not a Speech or OpenAI key. Add this header:

Ocp-Apim-Subscription-Key: <your-key>

If you see 401 Unauthorized, the key is wrong or missing. Rotate keys if needed.

4. Verify Project Status

Your CLU project must be:

• Trained successfully (no failed runs).

Deployed to an active deployment name (e.g., prod or staging). Runtime calls fail if the project is not published or deployment name is incorrect.

5.Correct Request Body

Example JSON:

{
  "kind": "Conversation",
  "analysisInput": {
    "conversationItem": {
      "id": "1",
      "text": "Bonjour, je veux annuler ma réservation",
      "language": "fr"
    }
  },
  "parameters": {
    "projectName": "<your-project>",
    "deploymentName": "<your-deployment>",
    "verbose": true
  }
}

404: Wrong path, API version, or project/deployment name.

9. Networking and CORS

If public access is disabled or IP restrictions apply, browser calls will fail. Use a backend proxy to keep keys secure and avoid CORS issues.

10. Checklist

• Endpoint matches resource region.
• Correct api-version.
• Valid subscription key.
• Project trained and deployed.
• Request body schema correct.
• Networking allows access.