In the time of the pandemic, all age group of students are studying virtual. We can see all of the students studying from home. Because of that, they are giving exams online and their writing habits are fading aways which is the most important thing in childhood study.
What it does
It is all in one study platform which can be adopted by schools directly. First of all students and teachers need to make an account and by using SQL query from the google cloud platform it will check the credential and then will allow students and teachers to sign up/log in.
If the users are students then with credentials they can log in to the system and they will be redirected to the calendar which is built using google calendar API. From that, they can directly see the assignments due, completed etc. Then if some student wants to complete the assignment then they can easily download the assignment from the app and take a print out of that and manually fill it within the given boxes and then upload the scanned answer sheets. We have used the OpenCV library in python which firstly detects the boxes on the page then crops those boxes out then passes those images as a list then another handwriting recognition model from the back end and passes the recognized alphanumerical values as a dictionary with the previous image list and checks that with given answerkey taken from the teacher(another dictionary) and calculates the marks. This way using multiple ml models makes the work of teachers easier so they can more focus on studies rather than taking the exams and manually checking all of the papers and if there is a mistake in checking then there is an option of the flag through which teacher manually check the paper and give the marks for the rare cases.
If a teacher logs in to the systems then they have the analysis option through which based on the marks they can check the progress of student studies as well as class growth with some graphs. There is also a face gender and age detector model implemented within the system. Students need to give the exam with their camera on (If a teacher wants to) and if the camera detects another face rather than a student then it takes the screenshot and alerts the teacher as well as parents with email.
How we built it
We have built it using a different tech stack. We have made the android app and we have made face, gender and age detector ml model as well as handwriting recognition model using python, TensorFlow and Keras. We are storing, saving and corresponding to every kind of data from google cloud.
Challenges we ran into
In the initial stage, the machine learning models were trained on small datasets so they were having high accuracy but not accurate results but then we found some open big datasets, retrained our models and increased the accuracies of all ml models.
Accomplishments that we're proud of
We were able to make the models with more than 89% accuracy and we have created such a beautiful interface too. We are proud that we were able to test the dummy data with google cloud and we succeeded.
What we learned
We have learnt a lot of things from designing to machine learning to project management and how to work under pressure.
What's next for Study buddy
In the next Study, buddy will be published on google play store and it will also have one extra feature of custom exam time allocation so the teacher can schedule exams and they can also proctor the exams if they want in real time. We are also working on the video calling feature so soon that also will be available.