Structured Output

One faster way would be to customize the prompt about how you want the output to look like. But it is not guaranteed whether it will work all the time or not.

Other way is to check Structured Output tools available in Llamaindex: https://gpt-index.readthedocs.io/en/stable/core_modules/query_modules/structured_outputs/root.html#structured-outputs

@WhiteFang_Jr Thanks a lot for your answer. So i am doing with a prompt and the output is already a schema, but this does not help in case the context limit is reached for a prompt

Do try the above stuctured output tool. They will help you to avoid token limit

I will try. Thank you

And one more thing is does our index get more intelligent as we store more chat and data

I will try structure output tool and update you

Umm no, it does not get intelligent per se. But if you are adding new data, it'll definately be able to cater new areas of query that comes with the new data plus if you improve previously stored data then also there is an impact on response generation part.

What if we index the summary of chat

We will add it to the embeded collection

That will only improve the index quality if the response returned by the index was initially correct.

Thanks

I have 3 indexes each containing quater result of a company, how do i create a chat enginer

I am thinking of graph

Is it a good approach