Redu | Devpost

Inspiration

A huge barrier to entry for underrepresented communities and active learners is the inaccessibility of finding key information from textbooks and large amount of text. Even if textbooks are open to the public, which many are not, users cannot find the information they are looking for.

If users can access textbook information in constant time, it will allow for democratization of information - allowing under represented groups to learn, students to learn from beginner to expert research papers, and continuous learners to continue growing.

What it does

We are aiming to redefine education by allowing users to ask our bot a question, and we will pull up the exact sentence from a textbook with the answer. After, you can interact with our bot to ask more questions about the text - trained exactly on that data. If the chatbot cannot answer the question, our discussion form will allow experts and beginners to communicate by selected topic. Finally, the platform will work in multiple languages.

How we built it

We used React to build our front end and we hosted our model using amazons s3 instance for Flask. To implement the semantic search on the textbooks, we used np net transformer model from hugging face, which we fine-tuned it to embed it in our application and ensure accurate. Afterwards, we enlarged the search queries using the open-ai api. Additionally, the chatbot used open ai api, and on the enlarged search queries, we ran the nearest neighbor algorithm.Finally, the user database is personal database we made it from scratch on top of sql and JWT tokens were used for all user and database auth. The discussion section was done in is react js, with no external libraries, all from scractch.

Challenges we ran into

Our server was giving a CORS error as the server was not secure so it cannot access the port. We opened the port with jwt tokens to fix this.

Accomplishments that we're proud of

Being the worlds first semantic search for textbooks using inferance is something our team is really proud of! Finetuning our model on textbooks and having it pull up the exact sentence will revolutionize education.

What we learned

We learned how to finetune an embedding model for textbook semantic search, and pull up the exact sentence. Additionally we are proud of being able to provide quick inference of the material and hollistic learning.