Pineroot
Explore research efficiently.
The problem Pineroot solves
As student researchers, we often have to independently learn advanced topics at a rapid pace, with little assistance or guidance. But with dozens of citations per paper and topics that we are vaguely familiar with, we are often confused on what reference we should look to next to best improve our understanding of the topic. We might pick a paper based on a friend’s interest, or maybe pick a reference solely based on its author or title. Existing LLM-based methods have the potential to hallucinate false information and nonexistent papers. We hope to save valuable time and provide more relevant results by pointing researchers to research papers that are most critical to further their understanding on the topic they are exploring. Instead of having to blindly guess what reference they should look at next, our application uses RAG to analyze all relevant citations of the current paper and succinctly explains to the user which papers are the most relevant to their interests in an intuitive interface.
Challenges we ran into
Backend
There's a lot of different APIs out there for getting research papers using Python. Go figure. Because of that, and the limited functionality and integration between each of them, we had to make a decision early on -- without perfect information. As a result, we found ourselves pivoting between a bunch of different APIs from ArXiv, Scholarly, CrossRefs, etc... which definitely took a toll on us.
Frontend
Going into the hackathon, all of us had limited frontend and React experience. Despite that, we ended up settling on an idea that was very frontend-oriented and UI/UX heavy. In addition to that, we decided to use a library that was ill-maintained and lacked documentation in a lot of key areas that were important for our visualization. This mixed together to form a ton of different roadblocks at every given turn we took.
Aside from this, we didn't realize this at the time, but making a tree type UI that looks good and performs well is surprisingly difficult. So with that in mind, we struggled to find any way to execute our vision for what it should look like and perform the way we wanted it to.
Sleep Deprivation
As with every hackathon, sleep was a huge issue. 2 of our team members (Akash and Kevin) both slept in Klaus directly. This, to say the least, hampered their productivity.
Tracks Applied (2)
Best Use of MongoDB Atlas
Major League Hacking
Generative AI - Presented By Microsoft
Technologies used