RedPen AI

AI-powered exam grading using Gemini 3 - automated correction, student feedback, and class-level insights in minutes.

Comment

Inspiration

The idea for RedPen AI comes from a real classroom experience.
When I was in middle school, one of my teachers once told us that grading exams was the hardest part of his job. He was responsible for several classes — more than a hundred students — and correcting exams always took days, sometimes weeks. He even joked that if a machine could grade exams, teaching would become much easier. Years later, with the rise of modern AI systems, I realized that this “joke” had become a realistic problem to solve. That is how RedPen AI was born.

What it does

RedPen AI is a web platform that uses Google Gemini 3 to assist teachers with exam grading.

Instead of manually correcting dozens or hundreds of exam papers, a teacher can:

  • Upload a correction key (questions, grading rules, scoring criteria)
  • Upload multiple student exam papers
  • Let Gemini 3 analyze each submission using reasoning, not simple text matching

The platform then generates:

  • A personalized correction report for each student, including:
    • Final score
    • Correct and incorrect answers
    • Explanations and feedback
    • Identified weaknesses
    • Suggested learning resources (e.g. YouTube videos)
  • A class-level report for teachers, including:
    • Common mistakes
    • Shared weak concepts
    • Students who may need extra support
    • Pedagogical suggestions to improve learning outcomes

How we built it

The project was built using Google AI Studio with direct integration of the Gemini 3 API.

Key design choices:

  • Gemini 3 is used for reasoning-based grading, not keyword comparison
  • Each student is evaluated independently and fairly
  • Reports are generated in a structured and readable format
  • The UI is intentionally simple to keep the focus on clarity and usability The application workflow:
  • Upload correction key
  • Upload student papers
  • Start grading session
  • Generate student reports
  • Generate class report

Challenges we ran into

  • Designing prompts that enforce consistent and fair grading
  • Ensuring Gemini 3 follows grading rules strictly
  • Balancing simplicity of the interface with meaningful insights
  • Testing without access to a real exam dataset (I generated realistic test data myself)

These challenges helped me better understand how to guide large language models for structured evaluation tasks, not just free-form text generation.

Accomplishments that we're proud of

  • Built a fully functional AI-powered grading platform
  • Successfully integrated Gemini 3 for reasoning-based exam evaluation, not simple text comparison
  • Designed a system that generates both individual student reports and a class-level analytics report
  • Implemented structured, fair, and consistent grading using a correction key and clear evaluation rules
  • Demonstrated that complex educational workflows can be handled using Google AI Studio
  • Created a clean and intuitive interface focused on usability for teachers
  • Tested the platform with realistic exam data to simulate real classroom conditions

What we learned

  • How to use Gemini 3 for complex reasoning tasks
  • Prompt engineering for evaluation and feedback
  • Building an end-to-end AI-powered workflow
  • Designing products for real-world educational constraints
  • Presenting AI systems clearly to non-technical users

What's next for RedPen AI

  • Teacher accounts with grading history and progress tracking
  • Advanced analytics dashboards for long-term class performance insights
  • Multilingual support to serve a broader global education community
  • Privacy and compliance improvements for real-world deployment

RedPen AI has the potential to evolve into a comprehensive AI assistant for educators, helping them save time while improving learning outcomes.

Built With

Share this project:

Updates