StudyBuddy Pro

Adaptive AI-Powered Learning Suite

Offline-capable smart quiz and flashcard generation from student notes using local NLP algorithms. A spaced repetition engine adjusts difficulty in real time. A gamified Pomodoro timer awards XP and coins, tracks daily streaks, and pairs with JWT-auth plus a local-first data architecture (browser IndexedDB for study content, SQLite for users).

Key features: Offline AI · RAG pipeline · Spaced repetition · Gamification · Local-first storage

---

What it does

Capability	How it works
Offline AI	Client-side tokenization, keyword extraction, and TF-IDF — no cloud LLM required for quizzes, flashcards, or summaries
RAG pipeline	Notes are chunked, indexed, and retrieved by relevance before generation so questions come from the strongest passages
Spaced repetition	SM-2 scheduling with Again / Hard / Good / Easy; due cards resurface automatically
Gamification	Pomodoro sessions, task completion, quiz scores, and review sessions earn coins and XP; daily streak tracking
Local-first	Notes, decks, cards, quizzes, and tasks live in IndexedDB (Dexie); user auth and rewards live in Express + SQLite with JWT

Architecture

React (port 3001)
  ├── Study NLP / RAG / SM-2  →  Dexie (IndexedDB)   [works offline]
  └── Auth + rewards          →  Express (port 5000) → SQLite users

• Browser: offline TF-IDF NLP, RAG retrieval, quiz/flashcard generation, SM-2, local persistence
• Express + SQLite: JWT register/login, XP/coins/streaks, profile settings

Features in the app

• Interactive dashboard (coins, XP, streak, level)
• Configurable Pomodoro timer with cumulative rewards
• Task manager persisted locally
• Smart quizzes & flashcards from pasted or saved notes
• Note summarizer (local NLP)
• Saved notes, decks, quiz history, and due-card review
• Theme toggle and profile settings
• Secure JWT authentication

Local development

Prerequisites

• Node.js 18+
• npm

Install

# Frontend
npm install

# Backend
cd studybuddy-pro-backend
npm install
cd ..

Backend environment

cd studybuddy-pro-backend
# Copy .env.example → .env and set at least:
# JWT_SECRET=a-strong-random-string
# PORT=5000

AI generation (on-device Flan-T5)

Quiz, flashcard, and summary generation use a lightweight local model in the browser:

• Model: Xenova/flan-t5-small via Transformers.js (~60MB quantized ONNX)
• Flow: TF-IDF RAG retrieves note chunks → Flan-T5 generates questions/summary
• Fallback: heuristic TF-IDF templates if the model fails to load
• No Gemini/OpenAI API key required for study generation

First run

The browser downloads the model once and caches it. Generation is slower the first time.

Optional: vendor model into the repo / GitHub Pages

npm run download-model

Files land in public/models/Xenova/flan-t5-small/.
public/models/ is gitignored by default (large binaries). To publish them on GitHub, use Git LFS for *.onnx.

Auth, Pomodoro rewards, and settings still use the Express backend.

Run

Terminal 1 — backend (auth + gamification):

cd studybuddy-pro-backend
node server.js

Terminal 2 — frontend:

npm start

Open http://localhost:3001 (production GitHub Pages builds use basename /studybuddy-pro).

Study tools (quiz, flashcards, summary, spaced repetition) keep working if the backend is stopped after the page has loaded. Login, Pomodoro rewards, and settings need the API.

Optional frontend env

REACT_APP_API_URL=http://127.0.0.1:5000

API surface

Method	Route	Auth	Purpose
POST	/api/register	No	Create user + JWT
POST	/api/login	No	Login + JWT
PUT	/api/user/gamification	Bearer	Add coins/XP; returns cumulative totals
POST	/api/pomodoro/complete	Bearer	+25 coins, +50 XP, streak; returns totals
GET/POST	/api/settings/:userId	Bearer	Profile JSON
POST	/api/summarize, /api/flashcards, /api/quiz	No	Legacy server generators (UI uses client NLP)

Deployment

• Frontend: GitHub Pages via npm run deploy (see homepage in package.json)
• Backend: Host Express + SQLite separately (VPS, Railway, Render, Netlify Functions, etc.) and set REACT_APP_API_URL for production builds