How abheytiwari/quizard works
Project type and purpose
• Quizard is a full-stack AI-powered PDF Q&A tool. Users upload a PDF via a Streamlit frontend, which calls a FastAPI backend. The backend extracts text with PyMuPDF, optionally computes embeddings, and issues queries to Google Gemini to generate answers.
File structure
• Top-level:
• README.md
• backend/
– main.py (FastAPI app entrypoint)
– embeddings.py (text vectorization logic)
– llm.py (wrapper for Google Gemini API calls)
– utils.py (PDF extraction, logging, config loaders)
– requirements.txt
• frontend/
– app.py (Streamlit UI)
• No separate data-persistence layer (embeddings likely in-memory or ephemeral).
• Deployment uses Uvicorn for the API and Streamlit’s server for the UI.
README for additional insights
• Tech stack: FastAPI + Uvicorn (backend), Streamlit (frontend), PyMuPDF for PDF parsing, Google Gemini as the LLM.
• Usage pattern: Upload → Extract → Embed → Query LLM → Return answer & logs in real time.
• Future plans hint at adding vector DBs, OCR, mobile UIs.
Elements
• Optionally, group related modules in a shaded “Backend” container.