Rully

Project Overview

This personal project aims to solve a recurring problem faced by Indonesian teachers. Despite advancements in AI and computer vision, many educators still rely on traditional methods to assess bubble sheets (answer sheets for multiple-choice questions).

To address this, I built an application that leverages computer vision to detect bubbles, identify corresponding letters, and determine which bubbles are marked or crossed. This core feature simplifies and automates the grading process for diverse bubble sheet templates.

Tech Stacks

• Front-end: Built with Next.js
• Back-end: Developed using FastAPI
• Server: Hosted on Render.com (free plan)
• Computer Vision: Used TensorFlow to implement a Convolutional Neural Network (CNN) model, specifically MobileNet

Limitation

• Server Downtime: Render.com’s free plan causes the server to spin down during inactivity, resulting in potential delays when restarting.
• No Database: The application currently processes data in-memory without persistent storage.

What New for Me

• How to build and train a computer vision model using MobileNet and TensorFlow.
• Integrating a machine learning model into a web application.