QuizTron

• 

  The quiztron homepage

• 

  Quiztron topic breakdown grid

• 

  A quiztron quiz, showing answers for review after submitting

Inspiration

What do you think you're an expert on?

There are many subjects which we pride ourselves on our skill and confidence. But how often is our knowledge put to the test on those exact topics? And how can we discover that maybe there are still things we need to learn?

On top of this idea that inspired the project, there was also the question of how AI will affect education, and how it already is causing problems for teachers who are trying to deal with AI essays and students using AI for cheating. Exploring the space of AI as an educational tool was also part of the inspiration behind Quiztron.

What it does

Quiztron is a website that lets you enter any topic your passionate about -- and then test your skills! Using Sonar Pro, Quiztron breaks down your topic into subdomains and concepts, and then drills you on those topics to find your weak spots, so that you can reinforce them.

Everything from academic subjects like Computer Science, Molecular Cell Biology, or Linear Algebra, to whatever you're passionate about: Harry Potter, 90s Sitcoms, etc. You can get quiz questions. Whether you're already an expert and want to engage with a topic you love, or whether you're just starting and want a fun way to practice and drill a new topic, Quiztron can help.

How we built it

Quiztron is built using python and FlaskAPI for the backend, Vue and PrimeVue for the frontend, with Supabase for authentication and storage. And of course Sonar Pro for generating the topic breakdowns and quiz questions!

Challenges we ran into

• LLM requests can be rather slow! They take a decent amount of time, and unless you stream everything everywhere, there are a lot of experiences where people are just waiting :P Streaming these into the UI would be a great improvement.
• As we know, LLMs aren't always accurate -- to say the least! Having LLMs responding in one way or another arbitrarily was difficult, and its essential to set up a playground in order to be able to tweak the prompts to return what's necessary.

Accomplishments that we're proud of

I genuinely enjoy playing with the site :) Even though there is definitely a lot of room for improvement, it's fun to answer questions on topics I'm knowledgeable about, or learn from the questions I get wrong. Or just click on a completely random topic and be humbled :P

What's next for QuizTron

There are many pieces that still need to be implemented or improved:

• Leaderboards, so you can see other people interested in the same topics as you
• The difficulty algorithm and ranking isn't working well, it suggests next difficulties that are too far at the extremes! And is bad at ranking when there are few results
• Currently the questions are per-subtopic; they would be better per-concept
• Implement a flow to report questions. We know LLMs don't always generate the right things! An option to challenge/flag/report a question is a must.
• LLM rating of freeform questions

Built With

• fastapi
• perplexity
• primevue
• python
• sonar
• supabase
• vue