Knovvu ML

It is further explained the configuration steps to use Generative Q&A feature through Knovvu ML. Generative Q&A can be used wİth Knovvu ML only, without the need of Knovvu VA. You can use Generative Q&A exclusively with Knovvu ML, eliminating the need for Knovvu VA. When RTG is sold as premium package and VA will not be used, this method can be used.

Configuration Steps

ML Configurations

1. Use {{identityUrl}}/connect/token method to get required access token. ClientId and clientSecret must be send in the request body.

2. Use {{baseUrl}}/sestekai-api/api/external/generative-question-answering-trainings/setting method to set openAI settings that is required for Gen&A.

   1. Enter the access token in the Authorization.
   2. Request body should contain the parameters below:
      • projectName: represents the project that will be used for GenQ&A by RTG.
      • provider: must be default "openai"
      • apiKey: must be provided by SESTEK or the customer.
      • model: must be default "gpt-4o-mini"
      • embeddingName: must be default "text-embedding-3-large"

   Example request:

   {
       "projectName": "Generative-QA-RTG",
       "provider": "openai",
       "apiKey": "sk-proj-g61YdoY3y7Bc5111111111111yWyx4uUPqRXX0ICVQKyyyyyyyyyyy8oT3BlbkFJSlvy4VtW2222222222gLxsGl0WA3i0Fi3s3rStUtQi_D-XE08YW589xxxxxxxxx",
       "model": "gpt-4o-mini",
       "embeddingName":"text-embedding-3-large"
   }
   

   Example response:

   {
       "projectName": "Generative-QA-RTG",
       "provider": "openai",
       "apiKey": "CnWPqlDrMDYFdh00000000000OPfO8oPSykngGJAFgmKxPlKgvZqqqqqqqqqqqqk0urJg0qm9Mx4LUyBmVxvPLVAEqqqqqqqqqqqqqMyv+GfT7Hg0ozwCr2f9cNJZ6GCzzzzzzzzzzzzzzz9jhFfnx",
       "model": "gpt-4o-mini",
       "embeddingName": "text-embedding-3-large",
       "embeddingDeployment": null,
       "deploymentName": null,
       "resourceName": null,
       "isDeleted": false,
       "deleterId": null,
       "deletionTime": null,
       "lastModificationTime": null,
       "lastModifierId": null,
       "creationTime": "2025-06-16T13:04:07.0153682Z",
       "creatorId": null,
       "id": "1d66ats6-3a0f-a9d7-25c1-3p6k8b03ostb"
   }
   

   The "id" parameter in the response represents modelId and will be used in the next steps.

3. Use {{baseUrl}}/sestekai-api/api/external/generative-question-answering-trainings/train to add required documents and train the model. After this step the project must be visible in ML interface.

   1. Enter the access token in the Authorization.
   2. Request body should contain the parameters below:
      • modelId: The model ID received in the previous step should be enter here.
      • projectName: must be same as the previous steps.
      • dataFiles: The files that will be used in GenQ&A feature should be uploaded.
      • dataUrls: The links that will be used in GenQ&A feature should be entered here.

4. Login to related Knovvu ML environment. You should be seeing the project you created in Projects page.
   

5. Enter the project you created. In the Experiments page of the project, you should be seeing the progress as Completed - 100%. This step might take few minutes.
   

6. Go to Published Models page in Knovvu ML. You should be seeing a published model with the name Published_ExperimentTitle, ExperimentTitle seen in the previous step. The Status should be Running. This step might take few minutes.
   Take note of the Public Endpoint, since it will be used in the next steps.
   

CA Configurations

1. Go to related CA tenant, Administration > Settings > Real-time Guidance, in Generative Q&A the settings below should be filled:
   1. Service: must be Knovvu ML
   2. API address: must be the URL you have as Public Endpoint while configuring ML, instead of /docs at the end, it must be /v1/text/question-answering/inference.
      
   3. Identity base URL: should be the identity URL of relted ML environment, e.g. https://sandbox.identity.stai.demo.sestek.com
   4. Client ID: Client ID of GenAI AI module (AIModule_SestekAI_App)
   5. Client Secret: Client ID of GenAI AI module (1q2w3e*)
   6. Prompt message: must be the default prompt as "Use the following pieces of context to answer the user's question..." You can get the full prompt from support/product team.