The problem Don't Ignore Me solves

Our main motivation to build this project was to solve the problem of getting ignored by teachers/instructors. Teachers often tend to ignore redundant queries as managing a team, assignments and tons of queries is quite difficult. So as a result, this has created a channel of inefficient communication. Back to the days where classes were held offline, we had bulletin/notice boards which would contain all the useful information and one could easily refer it. But now when the operations are online, there no one stop place where you can get information about notices, assignments and answers to your queries. Even Google Classroom has a limitation to that. Student’s redundant queries get ignores and notices are usually sent via mail.
Thus, to solve this problem of inefficient communication, we present Don’t Ignore Me.

Don’t Ignore Me is a portal where team admins/teachers/organizers can put up notices, assignments, alerts and setup a dynamic FAQ section. This portal brings the old-school feel of referring a bulletin board with a lot more functionality. From a member point of view you can get alerts about all the events and notices with a timeline for activities and access to an FAQ section where you can ask your queries. If the query is redundant, i.e it has already been asked before by the admin, you’ll be shown the answer or if it hasn’t been answered, the admin will be alerted. At the heart of the FAQ section is a Siamese LSTM Neural Network which is used to check semantic similarity between the questions and similar questions are shown only once in the FAQ section. This way the admin won’t be spammed for redundant questions and he/shes has to answer such question only once! The questions in the FAQ section appear in the order as most frequently asked question first.
This problem discussed above might not seem big enough but the impact of this solution will be very significant, reflecting in day to day operations.

Challenges we ran into

The biggest challenge we ran into was hosting the keras model with django. As a sufficient and comprehensive solution to this problem was not available online, we had to spend quite some time figuring it out ourselves as it was something we did for the first time.
The training of SiameseLSTM was also a big hurdle as it required a good GPU utilization and took around 8 hours to train on a Google Colab notebook.
Also finding a good and effective way to represent and organise the information was not an easy task as UX of the product is as important as the service of the product.