ResearchRadar.ai

Your research. Now, with AI.

Comment

Our Inspiration

Every week, researchers invest significant time staying updated with the latest scientific publications. The painstaking process of searching for relevant papers diverts valuable time from their core research activities, potentially hindering scientific advancement. Over a year, this translates into considerable time and financial inefficiencies.

We believe that by leveraging artificial intelligence, our project can profoundly enhance the efficiency of this process. We have developed an AI-powered recommendation system that adeptly identifies and suggests high-quality, pertinent research papers.

Moreover, our system customizes its recommendations by analyzing each researcher's reading history, guaranteeing both relevance and significance. Notably, our platform delivers recommendations on a per-project basis. This means that even if a researcher is juggling multiple projects simultaneously, our recommendations remain distinct and organized, preventing any overlap or confusion.

What it does

ResearchRadar.ai's features:

  • Using our AI-powered recommendation system, we help users find high-quality relevant papers
  • After a user finishes reading a paper, our system will recommend new research papers
  • Search papers with AI reordering using Metaphor’s API
  • Save research papers to a reading list
  • See important keywords for each paper based on the YAKE Python library
  • Annotate and view PDFs
  • Create multiple projects to maintain an organized reading experience

How we built it

AI-powered Recommendation system: We utilized the state-of-the-art “Neural Collaborative Filtering” algorithm through a Python library, which extends matrix factorization using a multi-layer perceptron. This approach enables us to recommend papers that align with the broader consensus of the research community.

In addition, we used the following tech stack to build our product.

Frontend: JavaScript, Chakra UI, React.js, Next.js

Backend: Python, Metaphor API, Microsoft Recommender library, PyTorch, Flask, YAKE (Yet Another Keyword Extractor Python library), Git

Challenges we ran into

While building ResearchRadar.ai, we ran into a few technical challenges, but we resolved them after ample brainstorming among ourselves and mentorship from the hackathon experts. Firstly, we needed to rapidly learn new technologies without having previous exposure. For example, some of us did not have experience working with Flask or Chakra UI, but we were able to pick up these technologies quickly.

Secondly, a major challenge we encountered was integrating the frontend and backend together. Specifically, we found it difficult to implement the machine learning model because it took a fair amount of time to run the model, which led us to asynchronously running the ML model, causing a some syncronization issues.

Accomplishments that we're proud of

Our group is extremely proud of the fact that we were able to get the recommender system to function properly. The ML aspects were technically challenging, and this feature served as the basis for our project. This feature was our MVP, and our hard work and dedication to the project was instrumental in achieving our project goal. Furthermore, when we successfully integrated the frontend and the backend together, we felt accomplished to present a working product for users to enjoy.

What we learned

We learned the common techniques and algorithms used to form the basis of recommender systems. In learning about the differences between content-based filtering versus collaborative filtering, we discussed the pros and cons of each one, and ultimately decided to implement collaborative filtering for our recommender system. We also learned about the flow of data and how data is managed between the frontend and backend portions.

What's next for ResearchRadar.ai

For future features, we hope to add a community feature to provide researchers the platform and exciting opportunity to discuss papers and voice their opinions. Additionally, we hope to add a feature that allows scientists to “tag” papers, thus providing the ability to categorize research papers by topics. We believe that our platform can be a one-stop shop for researchers to be able to catch-up on latest trends and further their knowledge.

Built With

Share this project:

Updates