The problem EduPack solves

EduPack is a complete educational package that makes textual learning interesting and easy. It provides audio and visualizations in the form of a video to make understanding and retention of concepts easier.
Have you ever sat down with your biology textbook and complained how mundane and tedious it is to skim over volumes of texts without anything getting registered in your brain? It has been proven that most people prefer a visual or audio way of learning/understanding things. However, it is difficult to find a relevant youtube video for what you’re studying. That is exactly the problem we have tried to solve with EduPack takes from the user a piece of text which text which is difficult to memorise and returns a video lecture which you can see at your own ease. The video lecture consists of relevant images, diagrams, structures, charts and graphs needed for better understanding of the topic. Along with that, you no longer have to read the text! We have used a text to speech converter, which means there will be a voice reading out everything. All of you have to do is, sit back and learn.
An additional feature is our auto-generated questions and answers. If you’ve just finished reading some topic, you usually look around for sample questions and see if you can solve them. Sometimes sample questions even help in revising a particular topic. You can easily achieve all this with the help of our auto-generated questions set. For every piece of text, you provide to EduPack, it gives back a set of simple objective questions which you can answer within in minutes and estimate your knowledge. EduPack scores your answers and also tell you the correct answer for every question.
The major target audience of EduPack is school students who don’t always enjoy subjects like Biology, Geography, History and would love a better way of learning those subjects. Instead of purchasing expensive audiobooks or sample question sets, they can simply use EduPack!

Challenges we ran into

1. Separating the given text into sentences:
   Problem: We can’t simply slice the given image at full stops as they are used for a variety of purposes like indicating an abbreviation, decimal in numbers, etc. Some sentences also end in exclamations and question marks.
   How we solved: We used python’s nltk library for natural language processing. It has a tokenizer to tokenize text into sentences using its pre-trained models
2. Amount of time taken to generate and render the video:
   Problem: We used a library moviepy to create the video. However, the images scraped had varying and large file sizes. The frame rate also affects the time taken to render the video.
   How we solved: We used Google's advanced search settings to improve the type of images downloaded to suit our needs like limiting filesize, etc. We even decreased the framerate as the video is basically a sequence of images stitched together. It affected the smoothness of transition but that’s a tradeoff we have to increase rendering speed.
3. The relevance of content generated:
   Problem: Since, the images are scraped from google images, many times, the images fetched are quite random. Keyword generation sometimes turns out to be irrelevant from an educational point of view.
   How to solve: Create a dataset of educational images with proper tags which was beyond the scope of hackathon. Rather than using pre trained models trained on mixed text, we should train our custom model with dataset taken from educational texts. (Again beyond the scope of hackathon)
4. Variety of questions generated
   Problem: Currently, we are only generating fill in the blanks types of questions for demo purpose based on keywords.
   How to solve: NLP libraries provide part-of-speech tagging of words based on nouns, pronouns etc. These can be used to create more complex and useful questions, however we could not explore this during the hackathon.