It is a pain to read through massive articles/essays/other bodies of text, and most of the time it is quite hard to absorb all the information, or even just the important information. This prompted us to create MARA, a tool to solve both these problems.

What it does

The user inputs a txt file containing a body of text of their choice (e.g. article, essay, paragraph, etc.) and we use ML to summarize this body of text and return the key information. Our question generator tool also has the user input a txt file containing a body of text, but here instead of summarizing we use more ML to generate a list of questions and corresponding answers and return them to the user.

How we built it

We built MARA using several languages/libraries:

  • Python
  • Django
  • Pipeline/PyTorch
  • HTML/CSS/Bootstrap CSS
  • HuggingFace Transformers

Challenges we ran into

At first, we had a very ambitious vision for MARA. We wanted to have the user manually input (or copy and paste) a body of text instead of having them give us a txt file. We wanted to have the question generator portion not only generate questions, but also give the user a quiz. We also ran into issues when returning the generated questions back to the HTML file, and we had to spend some extra time fixing this.

Accomplishments that we're proud of

We're proud of getting a finished product out the door in time, and having it be quite polished. We are proud of mostly meeting our vision of what we wanted MARA to be, and also implementing several ML libraries in our app.

What we learned

We learned about Django, implementation of ML libraries in web development servers, passing information from Frontend to Backend through POST requests, and of course we also learned about the ML libraries we used (PyTorch, Pipeline).

What's next for MARA Parse

We would like to remove the txt file input and replace it with the aforementioned basic text input, and also have the question generator page give the user a multiple choice quiz based on their text. We would also like to improve the accuracy of our ML. We can also add a feature where users can turn video to text.

Built With

Share this project: