We all know that re-reading your notes is one of the most popular study methods amongst students, even though everybody knows it is ineffective. So why do we still do it? Because it's the easiest to do. We student need a solution that is quick, easy, and effective.
What it does
FlashRecall uses course notes and textbooks to generate flashcard questions using AI that test student's understanding of their material. This is a method known as 'Active Recall'; cited as one of the most effective ways to study. Once you highlight your problem areas on the course notes, FlashRecall can detect the highlight and ask you questions solely from your problem areas.
How we built it
We used Google's Vision API to convert input images to text. Then we generated three types of questions (define, fill-in-the-blanks, true/false) from the text using Google's Natural Language API and some basic search. The website was made using Flask, HTML, CSS and JS, and was deployed to pythonanywhere. And most importantly, we used a domain from domain.com.
Challenges we ran into
Detecting highlighted text by using bounding box operations was definitely challenging to get correct. Also, deploying to a platform was almost impossible given the amount of errors we faced. We learned a lot from these challenges :)
Accomplishments that we're proud of
We are proud of the idea itself because it is something we know a lot of students would find useful, including ourselves. Aside from that, creating questions from highlighted text turned out to be a super relevant feature since effective studying happens by focusing on your problem areas.
What we learned
We learned how amazingly accurate the Google Vision API's are. We tried to manually tweak parameters using pyTesseract before using Google's API's, and the difference in their versatility is huge. We also learned the process behind deploying apps, which we definitely underestimated at first.
What's next for FlashRecall
After making some tweaks to the question generation algorithm, I plan to actually use this myself. I think it'd work best for fact-based courses like Chemistry, History and Econ. If I find that it helps me study better, I could release it to the public.