Inspiration

As engineering students, my friends and I run into research papers that we need to summarize and understand. It has become easy to summarize papers with AI but we gain zero knowledge out of it. To get the best of both worlds, I thought of creating a text editor that requires you to have knowledge of the subject matter and only gives suggestions and facts from the paper.

What it does

This project allows you to upload a PDF file that is broken into chunks and vectorized. While typing, anytime you hit a colon ':', it takes the sentence before the colon as a prompt and gives suggestions with context from the PDF you uploaded.

How we built it

First we broke the PDF into chunks and vectorized the chunks and stored the vectors in a vectorstore. Then we built a frontend which is a basic text editor that can be used to activate the RAG pipeline through a colon. Then the query and context from the RAG model are sent to a LLM and the response is returned.

Challenges we ran into

We had to test multiple embedding models and LLM's to ensure we get an appropriate result. We had to fix multiple errors related to the integrating of the frontend and backend.

Accomplishments that we're proud of

We are just proud that this works as intended.

What we learned

We learnt to work with different API's, different types of embeddings, the overall working of a RAG model and to make it from scratch as well. We also learnt how to implement different models for different purposes and to maximize our prediction accuracy.

What's next for Typing Assistant

We plan on improving the project even after the submission and releasing it to our college. We want to upgrade the text editor to make it like a regular text editor.