DocXBud

An AI-powered study assistant that transforms your lecture notes into an interactive learning experience using a multi-node LangGraph agentic pipeline.

Demo

Watch Demo

Features

• Chat with your notes — Upload any PDF and ask questions grounded strictly in your document using multi-query RAG
• Flashcards — AI-generated flip cards covering key concepts from your material
• Quiz — MCQ quiz with instant feedback, scoring, and full answer review
• Summary — Structured document summary with overview, key concepts, topics, and important details
• Chat History — All conversations persisted and resumable across sessions
• Auth — JWT-based register/login with bcrypt password hashing

Tech Stack

Layer	Tech
Frontend	React + Vite + TailwindCSS
Backend	FastAPI (Python)
AI Agent	LangGraph + LangChain
Retrieval	Multi-query RAG with LLM query expansion
Embeddings	sentence-transformers (all-MiniLM-L6-v2)
Vector Store	FAISS
LLM	Groq (llama-3.1-8b-instant)
Database	MongoDB Atlas
Auth	JWT (python-jose + bcrypt)
Rate Limiting	slowapi

Architecture

Multi-query retrieval — instead of a single vector search, the system generates 3 variations of the user's question and retrieves chunks for each, then deduplicates and merges. This significantly improves cross-topic answers (e.g. "compare X and Y").

Running Locally

Prerequisites

• Python 3.11+
• Node.js 18+
• MongoDB Atlas account (free)
• Groq API key (free)

Backend

cd backend
python -m venv venv
venv\Scripts\activate  # Windows
source venv/bin/activate  # Mac/Linux
pip install -r requirements.txt
uvicorn main:app --reload

Frontend

cd frontend
npm install
npm run dev

Environment Variables

Create backend/.env:

GROQ_API_KEY=your_groq_api_key
MONGODB_URL=your_mongodb_connection_string
SECRET_KEY=your_jwt_secret_key

API Endpoints

Method	Endpoint	Description	Auth
POST	/auth/register	Create account	No
POST	/auth/login	Login	No
GET	/auth/me	Current user	No
POST	/upload	Upload PDF	Yes
GET	/documents	User's documents	Yes
POST	/ask	Chat with document	Yes
GET	/messages/{id}	Chat history	Yes
POST	/flashcards	Generate flashcards	Yes
POST	/quiz	Generate MCQ quiz	Yes
POST	/summary	Generate summary	Yes

Key Technical Decisions

• LangGraph over simple LLM calls — stateful agent graph enables clean conditional routing across 4 modes with shared state
• Multi-query RAG — LLM-generated query expansion improves retrieval for comparative and cross-topic questions
• FAISS over cloud vector DB — lightweight, runs locally, no additional service dependency for the vector store
• Groq over OpenAI — fully free tier, no credit card required, fast inference suitable for real-time chat
• MongoDB Atlas — always-on free tier, flexible document model suits chat history storage