Socratic PDF

Inspiration

PDFs are common, like very common, but the way PDFs are usually consumed are in a passive manner. Users usually extract information from the PDFs and try to organize and construct a knowledge tree outside of the PDF viewer.

What it does

Socratic PDF allows users to interact with their PDFs (in this project, I specifically focused on being able to summarize, combine, contrast user highlights), making it easier for users to connect the dots with ideas throughout the paper.

How we built it

A lot of building a chrome plug-in is novel to me. I built the chrome plug-in following tutorials and of course, coding LLM :).

Challenges we ran into

My initial idea was really to allow LLMs run in the background, serving as a copilot, looking at relevant sources that users have highlighted and to have the LLMs read the relevant sources and try to integrate the knowledge from those sources into the context of my current reading. For example, if I am reading a paper about AlphaGo, I might also be interested in learning about how reference 7 that the paper uses provide more information to understand this paper. Previously, I would need to go off to read the paper in reference 7 before returning to the current paper and try to fit all the points in the reference 7 paper into the current context, however, if I had a copilot, I could send an LLM off to read the paper of reference 7 and have it figure out how that knowledge fits into the current paper. This approach would minimize the context switching and allow users to connect the dots more easily. This idea proved hard to execute because many PDF references cannot be easily retrieved, but is definitely theoretically possible. I pivoted to the current idea.

Accomplishments that we're proud of

Built a chrome plugin (never done before!). Was able to build a tool that I am proud to share to friends (though it might need more polishing before that :))