
Privacy-first AI study assistant. Offline Seq2Seq summarization, OCR with OpenCV, MCQ generation, 5 UIs in 1 codebase.
StudySage is a privacy-first AI study companion engineered to bridge local neural computing with web interfaces. Built to work seamlessly without cloud dependencies, it integrates offline Seq2Seq abstractive summarization, OCR via Tesseract and OpenCV, and interactive study tool generation across five distinct UI frontends.
Single core logic engine powering a single-port React SPA, Streamlit web app, CustomTkinter desktop GUI, Telegram bot daemon, and interactive CLI.
Runs local Seq2Seq transformer models (distilbart-cnn-12-6) for 100% offline document summarization without sending notes to remote servers.
Online mode uses Hugging Face Inference API for ultra-fast summarization of larger text snippets when internet connection is available.
Preprocesses screenshot captures using grayscaling, Gaussian blur denoising, Otsu thresholding, and morphological operations for maximum OCR accuracy.
Extracts clean text from lecture slides, scanned book pages, and screenshot clippings with automatic language detection.
Generates multiple-choice quizzes automatically using NLTK sentence tokenization and keyword frequency analysis with interactive scoring boards.

Clean, responsive React SPA with dark/light themes, OCR file upload, and real-time distilbart Seq2Seq summary generation.

Automated multiple-choice question generation with instant grading, explanations, and interactive scoreboards.

Lightweight cloud deployment on Streamlit Cloud with OCR processing, custom prompt parameters, and instant PDF downloads.

Native 100% offline desktop interface built with CustomTkinter for privacy-focused local study sessions.
$git clone https://github.com/sizwinz/StudySage-Offline-Online-AI-Note-Assistant.git && cd StudySage-Offline-Online-AI-Note-Assistant && python -m venv .venv$source .venv/bin/activate # On Windows: .venv\Scripts\activate
pip install -r requirements.txt$cd apps/web_app && npm install && npm run build && cd ../.. && python apps/api/server.pyOffline mode requires local PyTorch runtime. Online mode uses Hugging Face Inference API.
Compiles structured summaries and generated quizzes into publication-ready A4 PDF reports via ReportLab in Light or Obsidian Dark themes.
Built-in FastAPI server builds and serves the static React frontend and REST API on a single unified port (8000).
Native document loaders process raw .txt, .md, .pdf, .png, .jpg, .bmp, and .tiff files effortlessly.