Handwriting Recognition-Based Correct Answer Identifier

Clients

Techstack

Purpose

Grading stacks of paper exams by hand is slow and easy to get wrong, especially when a single test mixes handwritten answers with shaded multiple-choice bubbles. This system set out to score a whole answer sheet against its key in one pass, reading both kinds of answers the way a human grader would, only faster and more consistently.

Description

We built a desktop tool that photographs each answer sheet, identifies its format from a printed code, and isolates every answer region on the page. It transcribes the handwritten responses with a handwriting-recognition model and detects the shaded bubbles automatically, then compares everything against the answer key and exports the scored results to a spreadsheet. Reading messy human handwriting reliably enough to grade with is a genuinely hard recognition problem, and we packaged the whole thing into a tool a teacher can simply run.