Gradient

The all-in-one grading platform for teachers

HenHacks 2025 Winner

🏆1 of 10 major category winner groups out of 130 projects🏆 2nd Place 🥈 for Productivity Category!

Hosted by the University of Delaware

Sponsors: MLH | JP Morgan Chase | M&T Tech | Labware | Teach For America | Berkley | IWCS

How it Works

Gradient takes in two files: an answer key PDF and a student work PDF. These files can either be uploaded directly through your computer's finder or through a connected printer. Once the files are uploaded, our custom computer vision model (trained with Roboflow) segments the two PDFs into individual question PNG images. Two PNG images are produced for each question: the answer key question PNG and the student work question PNG. Using Gemini 2.0 Flash and its built-in OCR, we feed these two PNG images into the LLM and determine if there is a circling discrepancy between what was answered by the student and what was answered by the answer key. After determining whether or not the student got the question correct, we overlay "checkmark" and "x" PNG images onto a blank PDF and onto a marked-up PDF, both of which can be printed. You can either print out and view the marked-up PDF or you can print out and view just the overlay. If you wish to print the "check marks" and "xs" over the student’s work, you can do so by feeding into the printer the student's original work and printing on top of the paper the PDF overlay.

Step-By-Step Instructions

1. Clone GitHub repository either through bash or through GitHub desktop. For bash, enter git clone https://github.com/Klahadore/gradeint
2. Open the repository in an IDE or code editor.
3. Create a .env file containing your Google Gemini API key and your Roboflow API key.
   • Should be in the form ROBO_KEY='your_api_key' GOOGLE_API_KEY='your_api_key'
4. Open the terminal and run conda create -n gradient to create a VM.
5. Run conda activate gradient to activate VM.
6. Run conda install python (ensure python version is 3.13.2)
7. Run pip install -r requirements.txt to download all packages.
8. Run cd backend then uvicorn main:app --reload --port 8000.
9. Open another terminal and activate conda VM. Run cd frontend then python -m http.server 3000.
10. Copy the link http://[::]:3000/ into your browser.

Note

Feel free to use the two pdfs in the /example_pdfs folder and download them to your finder if you wish to use those files as the uploads.

Home Page

Loaing Page

Dashboard