Quizzically

Quizzically- Testing Simplified

Inspiration

As the pandemic hit, universities shifted entirely to online mode. We as first year (during that period), who already felt unknown to life in University of Toronto felt more lost than ever. One of the pandemic obstacles that stood out for us was the fear of giving online exams. Not only was their exam anxiety due to the uncertain/new format of paper, but also the recurring fear of loosing internet connectivity midway through the exam. These issues encouraged us to develop Quizzically, an offline software
interface that allows the professor to set up, conduct and proctor an offline exam (does not require internet connectivity).

What it does

Quizzically is an offline software interface that allows the professor to set up, conduct and proctor an offline exam (does not require internet connectivity). At the end of the exam, the software provides detailed question by question analysis of the average time spent by students on each question, as well as the Emotion/ response of students to each question using an AI model. This can help the professor determine the difficulty level of each question in the exam in relation to the response of the students, which can further help her/him compare the difficulty of the paper with previous years and mark accordingly.

How we built it

The team members used their diverse set of skills to come up with the final solution. The team has created the prototype using figma, the slide deck using Canva, while the AI models for student identification, emotion tracking and voice tracking have been developed, as follows:

Facial Recognition: Uses openCV to detect and track faces from the webcam. Can collect, train on, and recognize faces. Dependencies - openCV, numpy Run detect.py first to store 100 each images of subjects Run training.py to train models for facial recognition Run recog.py to detect faces.

Sentiment Analysis: Uses openCV, tensorflow, and keras to generate predictive models for sentiments and display emotions of person with webcam. Dependencies - openCV, tensorflow, keras, pandas, numpy Unzip dataset.zip into folder 'dataset' run trainer.py to generate new model, or use pretrained model.h5 run videoEmot.py to run sentiment analysis.

Voice-tracking: Records surrounding for a set period of time and generates transcript of spoken words for proctoring. Dependencies - pyaudio, speech-recognition, threading Run voice.py to start recording and transcription.'

Challenges we ran into

The team ran into several challenges while developing the project. One of the team members learned how to use Figma prototyping, while another team member trained the Sentiment Analysis Model from scratch for several hours. The video editing process was also time-taking due to some technical issues faced, but the team powered through with determination.

Accomplishments that we're proud of

The Sentiment Analysis model was developed from scratch and trained for hours, and finally attained 70-75% accuracy. This hackathon gave all the team members of Quizzically to play to their strengths, and further develop their knowledge in the area they worked on. The 24 hour time period was also a bit constrained for us, but we all managed to power through and complete the project in the given 24 hour period.

What we learned

We all left the hackathon with greater knowledge of the interface we were working with. The team members learnt how to embed animations in Figma, and a few team members were trying video editing for the first time. The brainstorming process allowed the team to research upon various diverse ideas, and thus gain knowledge regarding various new topics. We finally agreed on this idea, since not only would this make the digitization of learning and education more efficient, but Quizzically also provides a solution to a common obstacle we all faced at one point of time, therefore further inspiring us to develop this idea.

What's next for Quizzically

Our future goals for Quizzically are as follows:

The team would focus on further improving the accuracy of the AI models.
We also plan to implement compilers for Computer Science Assessments
The team would use AI models to assess discrepancy in understanding of topics in the future.