-
-
Logo
-
Front Page
-
Side Bar UI
Inspiration ⚡
Every studious student knows the importance of reviewing material in order to master and remember it. However, every student also knows that this is difficult and time consuming - not only do you need hours of concentrated study, you also need to make yourself the materials you need to study. In particular, we find that flashcards are a fun and effective way to learn, but often there is no existing deck of flashcards that matches your curriculum.
When we came to Hack The North 2023 and saw the amazing technology behind the Cohere LLM, we knew that it had the power to help. Ultimately, it gave us the idea for our hack: combine state-of-art machine learning models in a simple, and easy-to-use interface to allow students to study in a fun, streamlined manner.
What it does ⚙️
SmardCard can take notes from a variety of sources - including text files, textbook photos, screenshots , and websites - and generates a set of flashcards that test you on the key concepts from your notes. Your flashcards are completely personalized to you and your material, so you never have to worry about missing material when studying again! SmardCard also has a 'Quiz Me' option, that puts your knowledge to the test with AI-powered free-response comprehension questions, custom made for YOUR notes! It gives you the ability to generate a practice exam for yourself and have it graded by someone who knows your material.
How we built it 🔨
At the heart of SmardCard is the Cohere API, which gives us access to world-class LLMs that can be used for text generation. This technology powers both our flashcard generation and question generation/grading. The user's flashcards are stored on Cockroach DB, and the backend is implemented with Python and Flask. The frontend is built with NextJS, Typescript, and the Mantine component library.
Challenges we ran into 💀
The biggest challenge we ran into was the prompt engineering needed to create a reliable study companion, particularly in the 'Quiz Me' section of the app. Achieving an acceptable level of quality took a lot of prompt fine-tuning, writing examples, and tweaks to the generation system.
We also ran into a lot of issues implementing the backend, since most of our team members never did backend web development before. We had to learn SQL, Cockroach DB, Flask, REST, and more!
Accomplishments that we're proud of 🏆
Overall, we are really proud that we were able to complete our app, especially when remembering how much we had to learn throughout the process. We are most proud of our usage of Cohere's generative LLM, because it is really what powers the magical experience of having a computer teach you.
We also proud of the fact that it is being hosted on a website through tunneling, which is something else that we learned for the first time.
What we learned 🧠
We learned a lot about web development, particularly the back end. We also learned a lot about prompt engineering and generative LLMs, which required us a surprising amount of fine-tuning to achieve the optimal result.
What's next for SmardCard 📈
SmardCard still has a long way to go to become the ultimate study companion. First, we want to add more sources for notes, like handwriting recognition, google drive, and others. Second, we want to implement a mobile version of the frontend to streamline use of the app and allow students to study no matter where they are. Finally, we want to give students more study methods, such as AI-generated fill-in-the-blank worksheets to help students study the way they feel most comfortable.
Log in or sign up for Devpost to join the conversation.