The problem Proctor solves

Due to COVID-19, everything is being moved online. Unfotunately, this includes tests and exams, and students have been using this to gain an unfair advantage by googling answers or checking their notes. That's why we built Proctor.

Conventionally, you might think that the best way to conduct exams online is to ask students to share their microphone, webcam and screen to an invigilator. But this isn't without its caveats. Even in a class of 60 students, monitoring all of them at the same time is not a simple task. That's where Proctor comes in.

Proctor is a web app that facilitates the setting of exams and is an all-in-one platform to conduct exams online. With Proctor, you can upload questions, and when students take a test, they are monitored in real-time. If the model detects that the student is looking elsewhere or is no longer in frame, the invigilator is instantly notified.

What's next for Proctor?

Due to the time-constraints of the hackathon, we were able not add everything that was planned (especially a more polished UI). We plan to continue developing the project even after the end of the hackathon so that schools an universities around the world can use our system.

Challenges we ran into

We had no front-end people. Like seriously. All of us were backend based, and we struggled to build a clean-looking frontend. We wanted to incorporate a lot more features, but the development of the front-end could simply not keep up with the backend.

We also faced tons of hurdles when implementing the face detection. For unknown reasons, our ML guy @iagboola ghosted us, but in the end, we sorta figured it out with intense googling. Kudos to @pushpendrahpx for his enormous effort on the part.

@pushpendrahpx reporting, let me guide you through the intricacies of cheat detection.

The biggest issue we face was detecting the variety of ways a student could cheat. For instance, they could switch tabs to google something, or maybe refer notes on a side. Detecting all these edge cases was quite tricky, but we settled on using a model for a little more accuracy. We built the prototype using the Tensorflow body segmentation model which we then use to track the movement of the student.