Hi Pablo, thanks for putting all your efforts for such a great application development. Based on this git-hub solution we are trying do more with cognitive service and GPT3.5 and targeting to crawl all emails and attachment of any file type for current logged in and consent user, so that the end user can get into a conversation with in his mail boxes data. We are building a page and components for user authentication and targeting this application to be deployed as Teams app. But during working with the solution and services I have noticed few things that I would like to clarify with you. Please provide your feedback.
- I often get service unavailable or "SyntaxError: Unexpected non-whitespace character after JSON at position 6" error while even the index is not too big (15MB)
- what about OCR and extracted key phrases ? how can I make the GPT work on other fields except the content field. Adding these fields to semantic sometimes causes the toke error as well
- Merged content is too lengthy that mostly gives token limit error.
- directly asking a question out of a PPT / Excel or document is not always returning results like compare two different files , figures etc, Prompting relevant questions related to a document and after taking the conversation to that direction some times works nicely. But asking ad-hoc question from different documents stuck the application to generating answers and finally 504 error.
- What if the multiple instances of this application runs by different user , does the GPT3.5 model can handle multiple API requests or firmly say load balance. Do we need orchestration then ?
Microsoft
