Secure Handling of Consecutive API Requests

API
1

I am developing a CRM application focused on insurance brokers, which includes a module for extracting relevant information from insurance proposals in PDF format. For this functionality, I have integrated OpenAI’s Assistant, and the API call flow follows the process below:

  1. Creation of a Vector_Store;
  2. Upload of the PDF file;
  3. Attachment of the file to the vector, specifying the ID of both;
  4. Creation of a new Thread, specifying the ID of the Vector_Store;
  5. Execution of the Thread, providing the IDs of the Vector_Store and the PDF file;
  6. API call to messages to obtain the response.

The system is working very well in the application’s beta version, currently being used by 15 users daily. However, I am concerned about scaling to a larger number of users after the official launch. My question is whether OpenAI’s system can handle a high volume of consecutive requests without mixing up the data, as it is crucial to avoid responses being sent to the wrong user, which could lead to the exposure of sensitive information.

I would like to know if there is a real risk of data intermingling between different requests, and if so, what best practices should be followed to prevent this issue.

2

Yes

Always.

That’s up to you and your infrastructure. Ideally you would have some sort of mapping for each user and their respective documents in the form of IDs

1 Like