Saving prompts in an Azure OpenAI chat webapp

Question

Saving prompts in an Azure OpenAI chat webapp

Rohit Jain 0

We've been able to successfully develop a webApp that references our own data using gpt3.5 to generate content based on it. The app works great and chat history for each individual user works. However, we'd like a way to save prompts across multiple users so that good prompts can be labeled with use cases, serve as starting points for new users, and can be iteratively improved, ideally with change tracking. Is there a way to do that within the app?

Pramod Valavala 20,656 Reputation points Microsoft Employee Moderator

2023-11-08T18:39:41.9266667+00:00

@Rohit Jain Just checking in to see if the response here help

1 answer

Your answer

Pramod Valavala 20,656 Reputation points Microsoft Employee Moderator

2023-11-08T18:39:41.9266667+00:00

@Rohit Jain Just checking in to see if the response here help

Answer 1

Pramod Valavala 20,656 Microsoft Employee Moderator

@Rohit Jain With the current implementation of the web app, the chat history contains all messages related to the chat and the prompts would have to simply be extracted from there.

Another option would involve modifying the API (here would be good) that sends the prompts to another service for further processing as required.

Rohit Jain 0 Reputation points

2023-11-20T21:13:02.3333333+00:00

Ideally we don't want to save all the prompts. We only want to save some of them. So the approach of intercepting all the prompts wouldn't work.
Pramod Valavala 20,656 Reputation points Microsoft Employee Moderator

2023-11-20T22:33:55.3866667+00:00

@Rohit Jain That should be fine too. Since you would control the service, you can use whatever logic you need to capture the prompts that you what to capture.

This is the benefit of not having an opinionated catch-all persistence built into the Azure Service, allowing you to build your own layer as required.

Hope that clears things up!

Share via

Saving prompts in an Azure OpenAI chat webapp

1 answer

Your answer