PaperScout

• 

  PaperScout Interface

• 

  Sample Output

Inspiration

We took inspiration from a stack exchange post asking for a paper recommender system based on papers they input. While this post is a little old, we thought with the power of modern technology we could make that happen. Using modern AI models, our team’s app development skills, and a little too much caffeine, we can make this happen!

Link to Original Post: https://academia.stackexchange.com/questions/66323/paper-recommender-system?rq=1

What it does

Our app takes 2-5 DOI numbers and returns the user 5 recommended papers they can read to get more insight and information on a topic. We check that those papers can be utilized with the APIs we use. If they can, we ask Meta’s LLaMA 4 Scout 17B (Instruct) model to get DOI numbers that overlap in those papers. We sort those by how many times they were cited and if there are 5 present those to the user through clickable cards. If not, we ask Meta’s LLaMA 4 Scout 17B (Instruct) model to give us additional papers that it recommends to the user.

How we built it

We utilized a variety of different technologies to build the project. We utilized React for the frontend and Python for the backend. In the backend we worked on pulling data from the front end to process that data, make the API calls we needed to, and return an error if anything went wrong.

Challenges we ran into

We experienced issues of overlooking some information that turned out to be vital later in the process. We had some issues with communication with some team members losing contact/understanding of what someone was doing. Integration of the front and backend posed a bit of a struggle.

Accomplishments that we're proud of

We are super proud that we were able to get the entire app done and have it functional in the limited time of this Hackathon. We are grateful and proud of every member of our team for stepping up and giving it their all to get this project in on time.

What we learned

Each member of our team came from different backgrounds and experience levels. We all took away something from this project, some more than others. Some of our team members before this project had never worked with APIs or Nextjs and learned a ton about how they can be utilized in an app. While some of our more experienced team members learned a lot from all the small issues that came up along the way like integrating the front and backend.

What's next for PaperScout

We had several stretch goals or other features we weren't able to get into the final app in the time frame. We had an idea to let the user export the recommended papers into a Google Doc, translate the recommended papers into a new language, summarize the papers, suggestions for practical applications beyond the material, and visualize information in a 'knowledge map'. In the future, the team or someone else can maybe develop these features into PaperScout.

Built With

• css
• fastapi
• groq
• nextjs
• opencitationsapi
• python
• typescript