26Tanishk/marksheet-extractor ? reverse-engineered prompt

Build me a small web API that can take a student marksheet as an image or PDF and turn it into clean structured JSON.

I want it to read the document with OCR, then use an AI model to organize the messy text into fields like student details, exam details, subjects, marks, totals, and result info. If something is missing or unclear, it should return null instead of guessing. Please also include a confidence score so I can see how reliable each part of the extraction is.

Make it easy to test with Swagger docs, and package it so it can run locally and in Docker. The main goal is to handle noisy, inconsistent marksheets without breaking, and to fail clearly when the file cannot be processed. If you need to check current docs for the OCR or AI setup, go ahead and look them up online.