EduProctor

Inspiration

The project was inspired by the need to make classrooms more interactive and measurable in the age of hybrid learning. Educators often struggle to gauge student engagement and focus, especially in remote or large classroom settings. We wanted to create a system that uses AI-driven facial and attention analytics to provide real-time insights into student participation.

What it does

EduProctor automatically detects students’ faces from classroom video feeds, tracks their presence across frames, and estimates their attention levels using head pose and gaze orientation. It visualizes these insights in a professor dashboard, showing video output, detected student snapshots, and attention percentages per student. The system effectively bridges AI vision and educational analytics for smarter classrooms.

How we built it

We built the backend using Flask, integrating InsightFace for face detection, recognition, and embedding extraction. A Next.js + TailwindCSS frontend provides an intuitive interface for uploading classroom videos and visualizing analytics. Snowflake serves as the data store for embeddings and student metadata, while OpenCV and NumPy handle frame-by-frame analysis. The app runs end-to-end locally, processing classroom recordings into real-time engagement insights.

Challenges we ran into

We faced major challenges in frame sampling and head pose estimation, which initially resulted in inaccurate attention scores. Integrating InsightFace with OpenCV for consistent tracking was also tricky due to model dependencies. Handling large video files efficiently and synchronizing real-time results with the frontend dashboard required careful optimization and API design.

Accomplishments that we're proud of

We’re proud of building a fully functional AI pipeline that detects faces, identifies students, and quantifies attention accurately. The seamless integration between the Flask backend, Next.js dashboard, and Snowflake database showcases robust full-stack coordination. Most importantly, we transformed a research-heavy concept into a real-world educational analytics tool within a short timeframe.

What we learned

We deepened our understanding of computer vision, facial embeddings, and pose estimation, as well as the challenges of running AI inference efficiently on videos. We also learned how to connect cloud data systems like Snowflake to local inference pipelines and how to design scalable APIs for machine learning workflows.

What's next for EduProctor

Next, we aim to enhance EduProctor by adding emotion recognition. We also plan to deploy the system to the cloud, allowing institutions to manage multiple classrooms securely. Ultimately, EduProctor could evolve into a complete AI-powered attendance and engagement monitoring platform for smart campuses.