Menarik

• 

• 

  Summarization and research papers found by Menarik

• 

  Feynman Bot answering questions based on the found research papers.

Inspiration

I was having trouble reading research papers and finding research papers for one of my AP classes, so I came up with the idea of making a app that easily does that for you!

What it does

Menarik is a all in one research paper platform that allows for easy accessibility and understanding of the research papers!

It supports features like:

• Searching algorithm based off of the arXiv research paper repository.
• Feynman Bot: An AI Chatbot with context of the research papers we find for you, that can answer all your questions!
  • Feynman Bot Summarization of the research papers to save you time and help with understanding of the research papers!

How we built it

The frontend of the app is built on streamlit. The backend works by using the arXiv api to fetch the top and most relevant research papers for the topic the user is interested in and uses natural language processing with NLTK and LangChain to summarize the research papers. Feynman Bot is built with LangChain and GPT to answer user's questions on the articles.

Challenges we ran into

Some of my API keys were leaked so that was an issue lol. But the main issue was giving Feynman Bot memory of the research papers' abstract so it has context and can answer the user's questions. I also had to use NLTK to create a effective summary of the abstract to feed into langchain so that the processing time takes less.

Accomplishments that we're proud of

I am happy that I was able to create this app in the time limit while learning how to use langchain and the arXiv API because I don't have a lot of experience using them for my other projects.

What we learned

I learned many concepts on prompt engineering using LangChain and thats the main thing I am taking away from this hackathon.

What's next for Menarik

Full summarization of the research papers. Right now we only support abstract summarization but I want to level it up to the entire article. I also want to include more research repositories as well so that I could recommend more research papers from a variety of sources. I also want to make a recommendation algorithm but for research papers, so if the user clicks on the link to read a research paper from the research papers searched page, I know that the user likes the concept of it. So I can recommend more research papers with similar ideas to the user.

Built With

• arxiv
• gpt
• langchain
• nltk
• python
• streamlit