-
-
Example Error Explanation
Inspiration
We’ve all been there—losing easy points on an assignment due to small, avoidable mistakes. Whether it’s forgetting to multiply an expression by 2 or leaving out a constant, these minor errors can be frustrating and costly. Recognizing how common this issue is among students, we set out to build Presubmit, a digital proofreader designed specifically for handwritten assignments. Our goal was to create an AI-powered tool that scans handwritten work, detects careless mistakes, and provides interactive feedback to help students submit error-free work with confidence.
What it does
Presubmit acts as a real-time proofreader for handwritten work. Users can scan their physical documents using their device’s camera, seamlessly crop and adjust ratios, and process multiple pages at once. Once uploaded, the system analyzes the handwriting, highlights potential errors, and provides interactive explanations for flagged mistakes. Users have the flexibility to accept or ignore the suggested corrections before finalizing their work. When satisfied, they can easily share or upload the corrected document as a polished PDF.
How we built it
To build Presubmit, we integrated Google Sign-In for seamless authentication, ensuring users can securely access and manage their scanned documents. The front-end interface was developed with SwiftUI, prioritizing a user-friendly experience for intuitive document scanning and feedback interaction. On the backend, we leveraged Python and Gemini to accurately convert handwritten equations into digital text, allowing for error detection. Our custom logic engine was then designed to analyze and compare expressions, flagging potential mistakes. It does this by first passing images through a multi-modal large language model to extract text, math logic, and figure information. This information is then passed to a foundational model which is able t o find mistakes that the user is made. This is finally combined to place mistakes on the output image. Future iterations will focus on cloud integration to provide users with more storage and accessibility options. We also set up a cloud run instance to run the backend running a docker container for simple deployment.
Challenges we ran into
One of the biggest challenges was ensuring high accuracy in recognizing handwritten math notation. Unlike standard OCR, mathematical expressions involve symbols, variables, and formatting nuances that make accurate parsing more complex. Another challenge was optimizing real-time error detection without slowing down the user experience. Finally, integrating interactive feedback into a mobile-friendly UI required balancing usability with detailed analytics, ensuring that flagged mistakes were informative rather than overwhelming.
Accomplishments that we're proud of
We are particularly proud of successfully developing a functional prototype that allows students to scan their handwritten work and receive interactive AI-powered feedback. Implementing Google Sign-In and authentication provided a seamless user experience, and our handwriting recognition pipeline successfully detects and flags common careless mistakes. Most importantly, we’ve built a tool that has the potential to help students improve their accuracy and confidence in math by catching errors before submission.
What we learned
We gained valuable experience in implementing secure authentication using Google OAuth 2.0. We learned how to configure Firebase Console, manage client IDs, and securely handle authentication credentials through Google Service Info, ensuring a safe and seamless sign-in experience for users. We also explored Google’s Gemini AI for image processing, which allowed us to analyze handwritten work and overlay interactable buttons on detected errors.
What's next for Presubmit
Moving forward, we plan to integrate Google Drive’s API, allowing students to automatically save and organize their corrected documents in designated cloud folders. Additionally, we aim to implement a sorting algorithm that categorizes files into folders based on user-chosen symbols, making organization effortless. Beyond document management, we see potential in expanding Presubmit’s capabilities to include conceptual error detection, further helping students improve not just their calculations but also their understanding. Eventually, we envision Presubmit becoming an essential tool for students and educators alike, enhancing the way handwritten work is reviewed and submitted.
Log in or sign up for Devpost to join the conversation.