Analyzing large (100M+) csv files of financial transaction data

Hi all, we are trying to analyze financial data using gpt3.5-turbo. Specifically, we have csv files of company’s transaction data that we’re uploading to chatGPT via API. The files get pretty close to the 500M limit in some cases. While the upload succeeds, it fails when asked to analyze (example questions: what is the total amount of the payments in month x or year y? how many payments are in the file? etc) because it seems the file is too large. What is the right implementation pattern for allowing chatGPT access to large, financial data files for analysis?

Thank you!