StudyBuddy

StudyBuddy is an intelligent, RAG-powered (Retrieval-Augmented Generation) chat application that allows users to upload PDF documents and ask questions about their content. Built with a React frontend and a Flask backend, StudyBuddy extracts text from your documents, creates vector embeddings for semantic search, and uses a Large Language Model to generate accurate answers with direct page citations.

Features

• PDF Uploads: Seamlessly upload PDF documents to the system.
• RAG Pipeline: Ask questions and get conversational answers based strictly on the uploaded document's content.
• Clickable Citations: AI responses include exact page numbers that, when clicked, open the PDF directly to the referenced page.
• Streaming Responses: Real-time answer generation for a smooth, ChatGPT-like experience.
• Chat History: A collapsible sidebar retains your previous chat sessions.
• Change Document: Easily swap out documents and start a fresh session context.
• Firebase Authentication: Secure login via Email/Password or Google Sign-In.
• User-Isolated Storage: Every user gets their own dedicated upload directory, ensuring complete privacy of documents between accounts.

Tech Stack

Frontend

• React + Vite (TypeScript)
• Tailwind CSS (Styling)
• shadcn/ui (Accessible UI components)
• Firebase Auth (User authentication)
• TanStack React Query (State and data fetching management)
• React Router (Navigation)

Backend

• Flask (Python Web Framework)
• Flask-CORS (Handling Cross-Origin Resource Sharing)
• PyMuPDF / fitz (PDF text extraction)
• Sentence Transformers (Generating vector embeddings)
• ChromaDB (Vector database for similarity search)
• Groq API (Llama-3.1-8b-instant for fast LLM inference)
• Firebase Admin SDK (Token verification and secure routing)

Getting Started

Prerequisites

• Node.js (v18+)
• Python (3.10+)
• Firebase Project (for Authentication)
• Groq API Key (for LLM generation)

1. Clone the repository

git clone https://github.com/M-V-RAGHUPATHI-SAI/StuddyBuddy.git
cd StuddyBuddy

2. Backend Setup

Configure the Python backend and install its dependencies.

cd backend
python -m venv venv
source venv/bin/activate  # On Windows: .\venv\Scripts\activate
pip install -r requirements.txt

Environment Variables: Create a .env file in the backend/ directory:

GROQ_API_KEY=your_groq_api_key_here
FIREBASE_SERVICE_ACCOUNT_KEY=path/to/your/firebase-serviceAccountKey.json

Note: You must generate a Firebase Service Account JSON key from your Firebase Console (Project Settings > Service Accounts) and place it in your local backend structure.

Run the Backend Server:

python app.py
# The server will start on http://127.0.0.1:5000

3. Frontend Setup

Install the React dependencies and link Firebase.

cd frontend
npm install

Environment Variables: Create a .env file in the frontend/ directory, using the credentials from your Firebase Console (Project Settings > General):

VITE_API_URL=http://127.0.0.1:5000

VITE_FIREBASE_API_KEY=your_firebase_api_key
VITE_FIREBASE_AUTH_DOMAIN=your_project_id.firebaseapp.com
VITE_FIREBASE_PROJECT_ID=your_project_id
VITE_FIREBASE_STORAGE_BUCKET=your_project_id.appspot.com
VITE_FIREBASE_MESSAGING_SENDER_ID=your_sender_id
VITE_FIREBASE_APP_ID=your_app_id

Run the Frontend Development Server:

npm run dev
# The application will be accessible at http://localhost:8080 or port specified by Vite

How It Works

1. Upload Phase: Authenticated users upload a PDF. Flow routes to /upload, the PDF gets saved into an isolated uploads/<uid>/ folder. Using PyMuPDF, the text is extracted page by page, embedded using Sentence Transformers, and stored internally using ChromaDB.
2. Conversation Phase: The user submits a query to /ask. ChromaDB retrieves the most semantically relevant text chunks. The context is built and fed to the Groq LLM via streaming.
3. Citations & Interface: When Groq returns a cited answer, the frontend parses the __CITATIONS__ block and maps them to clickable components. Clicking a chip natively accesses the backend's /pdf/<uid>/<filename> route.

License

MIT