We all know that review is the best way to learn. However, when we put it into practice. we may encounter the following problem? what exactly content I need to learn and review in order to meet my learning goal? how to do enough reviews with minimum time? how to know the quality of your review so that you can customize your review times to strengthen your memory? Memstar are here to help with state of art NLP technologies

What it does

Memstar is a community platform that could let people create and share their customized learning plan. First, people can create and customize their review plan for their learning materials based on the Forgetting Curve. Second, for each review, our app will generate different multiple-choice questions based on their learning materials by using the state of art text to text NLP technologies " T5 transformers". people can get feedback based on that. More importantly, people can share their review plan to the communities. Those people who have the same learning goal could use these plan as their guide. Since we have perfect learning and feedback mechanism, I believe a lot of people could benefit from it and share their learning plan with the community, which makes the community more active. As the community have more excellent learning plan, more people will use our app and join the community.

How we built it

Web Frontend

Next.js is an open-source React front-end development web framework that enables functionality such as server-side rendering and generating static websites for React-based web applications. We launched our project on the next.js framework and use continuous integration and continuous delivery on Google Firebase platform Hosting service.

Web Backend

We use serverless deployment on the Google Firebase platform, and make use of the Firebase Functions to support our service. Its automatic scalability improves the reliability of our website. We also use Firestore as our NoSQL database to support our service.

Machine Learning

We trained a supervised machine learning model on Google Colab to generate questions based on T5 transformer and Google cloud NLP API (entity analysis).

Challenges we ran into

The procedure for generating distractors for multiple-choice and making an HTTPS backend server that has its own domain and load balancing proxy on GCP instance really makes us struggle initially, but we overcome it eventually.

Accomplishments that we're proud of

I apply the knowledge that I learned in my NLP class to real-world application

What we learned

How to deploy a real-world web application

What's next for MEMStar

Generate other types of question like True/ False, add picture and video

Built With

  • chakra-ui
  • colab
  • firebase
  • google-cloud-domin
  • google-cloud-load-balencing
  • google-cloud-nlp-api
  • google-cloud-ssl-certificate
  • google-cloud-vm-instance
  • next.js
  • node.js
  • react
Share this project: