Inspiration
Grading subjective student assignments and giving helpful feedback is one of the hardest jobs in education. Teachers often have a time keeping the same standards across different evaluation guides while also giving personal advice. On the student side getting a single letter grade without a detailed explanation makes it hard for students to understand how to improve. Rubriq AI was made to close this gap by turning evaluation guides into a smart interactive grading and learning helper.
What it does
Rubriq AI is a system that automatically checks assignments using evaluation guides and gives grades and helpful suggestions in a few seconds.
- Custom Rubric Processing: Teachers or students can create their evaluation guides with different parts and weights.
- Detailed Evaluation: Uses Google Gemini to check the work against each part of the evaluation guide and gives JSON format results making sure the scores are the same every time.
- Visual Performance Dashboard: Shows the scores through charts, clear markers and easy-to-understand explanations.
- Identifying Strengths and Weaknesses: Clearly shows where the student did well and where they need to improve.
- Personalized Practice Task: Makes an exercise based on the students weak areas to help them get better.
How we built it
- Frontend and Dashboard: Made with Streamlit and special CSS to give it a dark theme, glass-like boxes and columns that adjust to different screen sizes.
- Data Visualization: Used Plotly Express to make color-coded bar charts that show how good the student is at each part of the evaluation.
- AI Output: Runs on the Google Gemini API (
google-genaiSDK) usinggemini-1.5-flashwith models to ensure that the output is in the right format for scores, feedback, strengths, weaknesses and practice questions. - Deployment: Available through Streamlit Community Cloud and set up to automatically update when changes are made in a GitHub repository.
Challenges we ran into
- Getting Data Right: Making sure that the large language models give consistent answers for complex evaluations needed careful planning using Pydantic models.
- Designing a Good Interface in Streamlit: Streamlit's default options can look basic. To make it look modern we had to use CSS to add nice gradients, clean card designs and better visual details.
- Setting Up the Website: Moving from running the app on a computer to making it work online needed more than tunnels (like Cloudflare or Colab). We had to create a cloud setup on Streamlit Cloud.
Accomplishments that we're proud of
- Managing to get JSON format results from Gemini for complex evaluation work.
- Building a system that mixes numbers and charts with real learning advice like strengths, weaknesses and practice questions.
- Making a tool into a well-designed visually attractive product with smooth user experience.
What we learned
- Got better at using the
google-genaiSDK and making sure the data is correct with Pydantic models. - Learned how to change the way Streamlit apps look and work beyond the options.
- Got better at putting apps quickly using GitHub and Streamlit Cloud.
What's next for Rubriq AI
- Supporting File Types: Expand to check PDFs, handwritten work using OCR and code files.
- Grading Students at Once: Let teachers upload a whole set of student work to be graded all at once.
- Connecting to School Systems: Make it work with school systems, like Canvas, Moodle and Google Classroom through standard connections.
Log in or sign up for Devpost to join the conversation.