The problem TEXAM solves

INTRODUCTION

With the advent of pandemics, the world has explored new possibilities, how technology can transform the daily lives of individuals during the lockdown. The examination system has also seen many changes in the last decade. The pen-paper tests transformed to on-premise CBT (Computer-based Test) (like Jee conducted by NTA) than during lockdown Online examinations. Many platforms have emerged during this period and had their own contributions to the society, but they are lacking in terms of either reliability (credibility of the test) or Usability (convenience of using). Most reliable softwares are difficult to set up and use and needs training for teachers and students for using the platforms and costs a lot. But easy to use platforms were less reliable and needs additional assessment to trust the results. And these platforms are costly and need human proctors/invigilators.

NMIMS Mettl Case Study:
It surfaced in a Recent NMIMS Exams where a Serious Breach of privacy was observed.
Proctors from Mettl have gone to the extent of asking for social media profiles and messaging students on WhatsApp and harassing them. They even got hold of names and mobile numbers of Students (mostly Female students) from the Government Id which students were required to show for verification before starting the Exam.

Our Solution

Our Platform is supposed to eradicate the need of human proctors and save costs (Human proctors and computation). Our platform will benefit from the endless potential of Artificial intelligence and computer vision. Our platform is contactless and will prevent cheating in online examinations and the spread of viruses in on-premise Computer Based Test, our robust and secure software will monitor the computer hardware and ensure the credibility of the test.
We have discussed and consulted many industry leaders about the platform. And have got many positive reviews and some Technological Challenges for the platform.

Challenges we ran into

• Though Artificial Intelligence has limitless potential we had encountered some challenges in making hand gestures detection and face detection work with each other.

We solved this problem by restricting the region in which the hand gesture should be shown. this region is a rectangle at the top left corner.

• The Facial Recognition we were using was false detecting faces in the background and was giving false results.

We didn't find any specific solution for this problem, but then we did some hit - and trials with the input parameters (Configs) of the face classifiers like

scaleFactor, minNeighbors, and minSize

• There were some minor problems too, (like cors configuration, requirements and env setup) in which Google Helped alot