EasyQuizzes

Generating 4 Flashcards based on the book : Python Crash Course (548 Pages)

Inspiration

Having used Quizzlet for most of my tests the hardest task for me is first reading through all my notes and then making every Flashcard myself (hoping I grasp the right idea and my answer on the other side is right). I used to waste lot of my time preparing to prepare for the test than actually studying. With all the new development in AI and with Open Source models catching up with the State-of-the-Art models. Its easier than ever to create study notes

What it does

EasyQuizzes automates the creation of study flashcards from your notes. You upload a PDF, choose a topic, and set the number of flashcards you want. The app then creates knowledge chunks from these PDFs stores them to ChromaDB's vector database. When the user asks for N flashcards it pulls the K(preset value) most relevant chunks and the llama3.2 model uses them to generate your flashcards.

How we built it

We used ChromaDB as our vector database for storing chunk embeddings. FastAPI was utilized for API handling and backend testing. Meta’s LLaMA 3.2-3B-preview powers the app, with deployment made easier through Groq. The backend is written in Python, while the frontend uses VanillaJS, HTML, and CSS

Challenges we ran into

As a non CS grad, I had some programming background with LLMs and RAG , but deploying these projects was really hard the entire first day was spent on resolving issues with colliding packages in JS and Python. We almost entirely gave up on the tech stack.

Accomplishments that we're proud of

e're proud the app works as intended, especially under a tight 3-day deadline. The successful implementation of OCR was also a big win.

What we learned

We learned the value of teamwork, leveraging each other's strengths and past experiences. Delegating tasks effectively based on expertise helped us push through challenges.

What's next for EasyQuizzes

Post-competition, we aim to integrate VLLMs like LLaMA 3.2-11B-vision or smaller models like LLAVA or Moondream to add features like converting videos and audios into flashcards. We also plan to incorporate a NoSQL database to store flashcards and eventually develop EasyQuizzes into a full-fledged app. The current implementation uses groq hosted models which lack the finetuning needed for such specific tasks. So, we hope to deploy our own LoRA models for better output or understanding OCR content better.