Open YoranSys opened 1 year ago
Hey @YoranSys I agree about the customized prompt. It will increase the accuracy of the response, I only will add that this need to be added with the multiple datasets established (#338 and #348), because the customized prompt will be different for one dataset to another. And both ideas will contribute at the end for the accuracy objective.
Thanks for the great feedback @YoranSys / @jignnsd! As mentioned in https://github.com/danswer-ai/danswer/issues/375, we're planning to add a feature we're currently calling "personas". This will allow you to define custom document sets + prompts.
This is a fairly large feature and we have a lot of other features already planned. I'd expect it to come out in pieces sometime in the next month or two. Will add more comments in this issue with any/all updates!
Many thanks @Weves ! This will be a great improvement! Also, just to mention that will be great that you can choose one or more "personas" / datasets in your search. So you can interact with one "personas" or 2 or 3 or all. Many thanks for such a great project.
+1
Any progress on this feature? I am really interested in Custom prompts. I know we have persona that somewhat does this.
But we need to think, if we can build prompt template like in Langchain where I can pass input dynamically and make prompt more complex. Not sure how we can implement it, but it will be nice to have.
Not exactly a problem, but an enhancement. We would like to add customized prompts for different companies to improve the accuracy of Language Model (LLM).
A new feature that allows users to add personalized prompts. This would involve creating an interface or method for users to input their own prompts, which would then be used to train the LLM. The goal is to improve the accuracy of the LLM by tailoring it to the specific language and terminology used within a company.
Expected Impact This feature would enhance the usability of Danswer by making it more adaptable to different business contexts. It would also potentially improve the accuracy of the AI responses, thereby increasing user trust in the system.
We look forward to your feedback on this feature request.