BreakDown

Inspiration

Reading long-winded textbooks and lectures notes can be a difficult way to learn material. Our app provides a way to summarize content in a graphical structure, so that you can see important concepts and definitions, and most importantly, how content relates together.

What it does

Our app will take a pdf file and use AI to summarize and extract key concepts and definitions, and link it together. Then, it displays this content in the form of a graph that you can click through in an organized fashion to learn about the summarized material.

How we built it

We used HTML, CSS, and pure javascript, utilizing the canvas element for all our graph displays and animations. For the backend, we used python for both the endpoints for retrieving summarized information and LLM summarization responses. We used hugging face spacy/en_core_web_sm for English pipeline optimized for CPU processing.

Challenges we ran into

We ran into many challenges, like fine tuning the model to perform well on certain kinds of information. We also had issues with the algorithm we used to display our graph in an ordered structure.

Accomplishments that we're proud of

We are proud of our ability to integrate multiple disjoint components into one app, and how we came together as a team to implement the idea.

What we learned

We learned a lot about the use of LLM's and fine tuning models, and about visualizing using canvas in javascript. We also learned more about the implementation and application of graph algorithms.

What's next for Breakdown

We want to further fine tune our models and experiment with getting better and more structured breakdowns and summarizations of content. We would also like to improve the animations of our frontend.

Built With

• ai
• css
• fastapi
• html
• javascript
• llm
• pypdf2
• python