Inspiration
Students spend a lot of time searching through notes, PDFs, web articles, pictures of textbook pages in an effort to help themselves create quizzes to study from. Instead of relying on this, or teachers to provide quizzes to help study, I wanted to use the power of AI and modern LLMs to allow students to use any PDF, image, text content, or web link to generate a quiz from the extracted material to help them study.
What it does
heyquiz takes text content from PDFs, images, plain text, or web articles and passes it on to a LLM which returns a quiz for the student to take to help them study and retain material. The quiz scores the user, provides context for answers, and let's students organize quizzes by topic.
How we built it
The app is built entirely in Bolt, with the front end powered by React and the backend on Supabase. We use pdfjs-dist & tesseract.js for text extraction, which is passed along to Supabase edge functions that forward and return the data from Gemini 1.5 Flash.
Challenges we ran into
Figuring out token optimization and prompt engineering. There was also considerable effort in debugging and figuring out how to properly extract content from a document. Not being a developer (although I've worked a long time with engineering teams so I understand the lingo) was challenging, but luckily the Bolt documentation helped, Claude could help provide better prompts, and there were a few supporting threads on the subreddit.
Accomplishments that we're proud of
Being able to offer PDF, image, text and link extraction all rolled in to one that is quick & seamless.
What we learned
How to setup on Supabase and prompt effectively.
What's next for heyquiz
Adding in billing support and a comprehensive referral system. We also want to add support for Youtube video quiz generation, Powerpoint, Word Docs, audio, and Google Slides/Docs integration.
Log in or sign up for Devpost to join the conversation.