We’ve been testing the retrieval process for single-page PDFs containing minimal content, specifically extracting person’s names, times, and dates. However, each test takes approximately 25 seconds to complete. Is there a way to improve and expedite the retrieval process for such simple data extraction tasks? We’re looking for suggestions to optimize the speed while maintaining accuracy. Your insights and recommendations would be greatly appreciated!
Without information about the techniques, endpoints, and models you are using, along with a bit more details, I have no guess if you are either doing something completely backwards or have already invented breakthrough techniques unparalleled in industry…
If time-sensitive, the biggest boost would be in making parallel calls: chunking the documentation into a blast of dozens of calls with a completion in seconds expected. You then just have the likely requirement of algorithmic de-duplication.