BookAI

Saved
Explore section
Library

Inspiration

Inspired by the rise of addictive short-form content, our group chose to turn this primarily harmful genre of social media into a self-help and growth tool through AI, another quick growing tool, making harmful short-form content into something that works for social good.

What it does

This project uses an Artificial Intelligence model to give information and key takeaways relating to each book that supports learning through spaced repetition and micro-learning, mixing this learning with a social aspect. This enables users to learn from hundreds of pages in a faster, interactive, and gradual way.

How we built it

The front end was written using HTML5, CSS, and Javascript, and the backend was written fully in python. This uses downloadable books from a free book library, parsing the book, and then turning each paragraph in the book into vector embeddings through the Ollama API, which enables the model to sort through and use pieces of the book quickly, efficiently, and as needed.

Challenges we ran into

Throughout this project, we had trouble downloading the necessary books and making that work with code. In addition, parsing the book was complicated as many of the files were not in a great format to embed, making an already resource intensive and time consuming process complicated and inefficient. In addition, choosing a model to embed took some time, as we had to weigh the pros and cons of using each. Problems with embeddings were the most impactful and detrimental as embeddings are very time consuming and each failed embedding was costly and wasted time. We had issues fully integrating the model with the front end because we ran out of time to fully train and complete the model, however we were able to generate short summaries of books using the model.

Accomplishments that we're proud of

Can be scaled up easily due to efficient and unproblematic code Although this project had problems, we are proud of the strong potential for connection between the frontend and AI backend, as it can be easily scaled up and is unproblematic. In addition, the vector embeddings enable the model to give accurate and helpful information about the book, which becomes resource efficient as embeddings can be easily saved and reused. In addition, this project is more than just a proof of concept, the features currently implemented would not need much change to create a fully functional and working project/website.

What we learned

How to create a user-enticing app that is not only useful but is interactive based off of social media Throughout this project, we improved our skills on developing a full-stack application, and creating a setup for a front-end and back-end connection. In addition, we learnt more about embeddings and their power to make an AI's context window smaller and more efficient.

What's next for BookAI

Our next steps would be to create systems for user accounts, along with highlighting the social media aspect of this project more through shared tracking features such as streaks and progress checks. In addition, systems to source and add more books to the system that can handle more new book downloads and embed them on demand would be crucial to making the AI efficient and powered. Finally, features that would entice users to continue with their learning journey would be crucial to growing this website and developing its future and uniqueness.