AiQuizzer

Another ai playground project. Use the OpenAI API to generate multiple choice questions with context from PDF files - stored in chromadb - to test your uni knowledge.

Prompt inspired by this repo. Dockerfile for the frontend taken from this gist.

To do

• Refactor backend (Move chromdb and openai logic to separate files)
• Add logging with loguru
• Function to automatically process new PDFs in chromadb while docker is running
• Gamification (e.g. points for correct answers, emoji awards for streaks)
• Error handling (e.g. if no questions are generated)
• Exclude headers and footers from PDFs (tricky)
• Add Qualitiy checks in question generation (Multiple API calls? Or use GPT 4?)
• Refactor frontend (Undestand svelte, clean repo, make UI less ugly)
• Add more question types (e.g. True/False, Fill in the blank)
• Make docker use Pipfile for backend?

Local Deployment

Pre-requisites

• Docker
• Docker Compose
• OpenAI API key

Setup

1. Clone the repo
2. Create a .env file in the ./backend directory with the following variables:

OPENAI_API_KEY=<your key>
TEXT_FILTER=<your filter>  # e.g. "University of Hanover; Pof. John Smith". These strings will be removed from the PDFs before they are sent to chromadb

3. Create a docs directory in the ./backend of the repo and add your PDF files to it. Example structure:

backend
├── docs
│   ├── folder1 # Name of the folders will be used as collection name in chromadb
│   │   ├── file1.pdf
│   │   ├── file2.pdf
│   │   └── file3.pdf
│   └── folder2
│       ├── file1.pdf
│       ├── file2.pdf

4. Run the following command in the root directory of the repo and wait for the containers to start:

docker compose up -d -f docker-compose.yml

5. Access the frontend at localhost:3000

Development

Backend

cd backend
pipenv install
pipenv shell
python app.py

Frontend

cd frontend
npm install
npm run dev