in:decks

Inspiration

Being a student can be exhausting, and who would rather spend hours after class making flashcards when NLP could do it for you? We take care of the entire flashcard generation process so students can focus on studying for their next exam!

What it does

in:decks leverages the Cohere API to automatically process and create index cards from user-inputted text.

How we built it

Web App Frontend: Vue.js
Web App Backend: Flask
Input text processing: Co:here API, Python
- Utilized the generate endpoint to create a question and answer for the index card
- Utilized the generate endpoint to summarize long user input into a more concise and direct prompt
Semantic Similarity Analysis: Python, PyTorch, and GloVe pertained embeddings
Users can 'like' an index card if the generation was useful, and we will add it to our training dataset to make our algorithm even more accurate

Challenges we ran into

Connecting the frontend to backend
Dependency and compatibility issues
Finding values for the generate API endpoint that result in the best output (temperature, k, p)
Finding appropriate data to train Co:here model
Selecting the appropriate semantic analysis model with given resources

Accomplishments that we're proud of

Created a functional demonstration within the time given
Each other ❤️

What we learned

How to use Co:here’s API for text generation and summarization
How to complete basic similarity comparison tasks without available training data
NLP is very cool!

What's next for in:decks

Enable speech input (ie. lecture audio) and file input (ie. PDF file or Word document)
Incentives for students to study through gamification (point systems, rewards, challenges)
Other question modes (ie. matching, multiple choice, true or false)
Acquire more training data to improve model
Allow users to edit an index card and then we will update that in the training dataset
Our current model does not take into account sequence, only co-occurrence
Implement a personal user profile
Host on a web domain