Sisense Intelligence - struggling with requests

[francoisvv]

I am having trouble with the setup of Sisense intelligence. I keep getting the same message when sending requests so not sure where I am missing anything.  I set it up to use my OpenAI account and hope someone can confirm if my parameters are correct.  I keep getting the message "Oh snap, I'm having trouble understanding your question. Please rephrase or try asking a different question." so not sure if it a setup or account issue. 

Just for reference, I am using the API key I created in "https://platform.openai.com/settings/organization/api-keys", not sure if I should use the Admin API key, although I have tried both with same result.  Also not sure if it is the GPT version I am specifying, so any advice on which one is the correct one I should use.

I have attached a screenshot of my parameters, so any advice would be greatly appreciated.

[francoisvv]

After some struggles, I managed to figure out that I had to add credit to my account as OpenAI charges for every request sent and received, so I was able to get a couple of responses. 

However, the requests are extremely slow and 99% of the questions you ask, just returns no result.  Does anyone perhaps know what is the latest version that is supported and also have some guidance on how you can get better results?  Requests like "Give me the average "ActiveDays" from the Users table" or even "What is the most popular Name in the "Users" table" just simply yield no results.  Requests like "what are the columns in the "Final_" table?" does give results but it does not create a table or widget if asked.  Anyone played around with this yet?

[DRay]

Hello @francoisvv,

Thank you for reaching out. I see your question hasn't gotten a response yet, so I'm asking internally to try and get you an answer. 

[DRay]

Hi @francoisvv,

Do you know what version of the model you are using, and have you reviewed this documentation? https://docs.sisense.com/main/SisenseLinux/genai.htm?tocpath=Sisense%20Intelligence%20(AI)%7C_____2#...

I don't know why it would be slow or you wouldn't be getting any results. If you view the Dev Tools in the browser do you see any errors or anything? 

[francoisvv]

Hey David,

The responses are a bit quicker now so I think that is sorted, the main that I have come to see is the way in which you data is structured in your elasticube determines the response capabilities.  Some features are still disabled so I assume that this will be updated with the next version.

[yaellev1]

Hi @francoisvv 
What features are you referring to?
When setting up your elasticube, (or live model, no difference), please make sure that you hide any irrelevant columns, give meaningful names to the columns and provide descriptions where the column name may be ambiguous with other columns or unclear.

[francoisvv]

Thanks a lot for this information!  After doing so, the responses seems to have increased the speed in which the responses was returned so thanks a lot!

I noticed in the release notes, that some features are still unavailable.  Do you perhaps have an idea of when the new set of features will be rolled out, specifically the recommendations?

[yaellev1]

Thats great! The feature is likely to reach GA in next release (2025.2) and will include additional improvements in the chat response as well as other improvements to the UX.
By recommendations, are you referring to the AI generated quick start questions? if you feel that is not performing  well, we would love to get your feedback

[francoisvv]

At the moment I'm not getting any quick start questions so I thought that it might only be available in the next release.  Is it something that I should enable somewhere, or will this roll out along with the other features in the next release?  Thanks a million!

[yaellev1]

Hi @francoisvv, please reach out to support, and we'll investigate. You should be seeing the recommendations when the AI assistant loads.