Hello,
I have an issue that is fundamentally compromising an app I am building.
I would be SO grateful if anyone knows how I could overcome it.
SUMMARY
In the app, I upload a collection of json files that include text fields to a vector store.
Then I ask a GPT assistant to locate the json file and compare the text in one field with the text in another, to determine if there is a match.
If there is a match, the aim is for the GPT to respond with the matched text.
The problem is that if I just have a few files to process it is fine, but more than around 10, and it starts mixing up the responses. IE the matched text from one prompt is sent back as the response to another.
I am using AsyncOpenAI, and make simultaneous async API calls using a different assistant each time, and all the files in the vector store have unique ids / file names. I’m using 4.o and 4.o mini.
I have tried giving the GPT strict rules not to use any file other than the one specified, but it seems instead to be randomly choosing data from any file in the vector store.
I need a way to ring fence and isolate each API call so that when I retrieve the responses / messages the responses and messages are always and only the correct ones.
Anyone else had a similar problem?
And if so, what is the solution?
Regards and all the best
Mike McDowall