🧠 StudyMind

AI-Powered Teaching Assistant — RAG Pipeline

Upload your PDF notes · Ask anything · Get instant AI-powered answers

---

📌 What is StudyMind?

StudyMind is a full-stack AI teaching assistant that lets students upload their PDF notes and ask natural language questions about them. It uses a RAG (Retrieval-Augmented Generation) pipeline to find the most relevant section of your notes and generate a precise, context-grounded answer — no hallucinations, no generic responses.

---

✨ Features

• 📄 PDF Upload — Upload any text-based PDF; text is extracted and chunked automatically
• 🔍 Semantic Search — Questions are matched to notes by meaning, not just keywords
• 🤖 AI Answers — LLaMA 3.3 70B (via Groq) generates clear, concise answers
• 💬 Chat Interface — Clean chat UI with message history and loading indicators
• ⚡ Fast Inference — Groq's LPU hardware delivers answers in seconds
• 🌐 Fully Deployed — Live on Vercel + Render, accessible from any device

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                        USER                                  │
│                    (Browser / Mobile)                        │
└───────────────────────┬─────────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────────┐
│              FRONTEND — React + Tailwind CSS                 │
│                  Deployed on Vercel                          │
│         studymindv2-xi.vercel.app                           │
└───────────────────────┬─────────────────────────────────────┘
                        │  HTTP (REST API)
                        ▼
┌─────────────────────────────────────────────────────────────┐
│              BACKEND — FastAPI + Python                      │
│                  Deployed on Render                          │
│         studymind-jqmn.onrender.com                         │
│                                                              │
│   POST /upload          POST /ask          GET /            │
│   ─────────────         ─────────          ──────           │
│   Extract text          Embed question     Health           │
│   Chunk (500w)          Cosine similarity  check            │
│   Embed chunks          Find best chunk                     │
│   Store in DB           Prompt LLaMA                        │
└──────┬────────────────────────┬────────────────────────────┘
       │                        │
       ▼                        ▼
┌──────────────┐      ┌─────────────────────────────────────┐
│  PostgreSQL  │      │         External APIs                │
│  (Render)    │      │                                      │
│              │      │  HuggingFace Inference API           │
│  chunks +    │      │  sentence-transformers/              │
│  embeddings  │      │  all-MiniLM-L6-v2                   │
│  (JSONB)     │      │  → 384-dim embeddings               │
└──────────────┘      │                                      │
                      │  Groq Cloud API                      │
                      │  llama-3.3-70b-versatile             │
                      │  → Answer generation                 │
                      └─────────────────────────────────────┘

---

🔬 How the RAG Pipeline Works

PDF Upload                          Question Asked
    │                                     │
    ▼                                     ▼
Extract Text                        Embed Question
(pdfplumber)                    (HuggingFace API)
    │                                     │
    ▼                                     ▼
Split into                          384-dim vector
500-word chunks                           │
    │                                     ▼
    ▼                             Cosine Similarity
Embed each chunk              vs all stored embeddings
(HuggingFace API)                         │
    │                                     ▼
    ▼                             Best matching chunk
Store in PostgreSQL                       │
as JSONB                                  ▼
                               Prompt = chunk + question
                                          │
                                          ▼
                               Groq LLaMA 3.3 70B
                                          │
                                          ▼
                                    Final Answer

---

🛠️ Tech Stack

Layer	Technology	Purpose
Frontend	React.js + Tailwind CSS	Chat UI, PDF upload, state management
Backend	FastAPI (Python)	REST API, PDF processing, RAG orchestration
Embeddings	HuggingFace — all-MiniLM-L6-v2	384-dim semantic embeddings for chunks & queries
LLM	Groq — llama-3.3-70b-versatile	Answer generation from retrieved context
Database	PostgreSQL 18	Stores text chunks + embeddings as JSONB
PDF Parsing	pdfplumber	Multi-page text extraction from PDF files
Similarity	NumPy cosine similarity	Finds most relevant chunk for each question
Frontend Deploy	Vercel	Auto-deploys React on every git push
Backend Deploy	Render	Persistent Python web service + managed PostgreSQL

---

📁 Project Structure

studymind/
├── backend/
│   ├── main.py              # FastAPI app — all routes & RAG logic
│   └── requirements.txt     # Python dependencies
│
├── frontend/
│   └── src/
│       ├── App.jsx          # Main React component — full chat UI
│       ├── App.css          # Global styles
│       └── index.js         # React entry point
│
└── README.md

---

🚀 Local Setup

Prerequisites

• Python 3.10+
• Node.js 18+
• PostgreSQL (local or cloud)
• HuggingFace API key → huggingface.co/settings/tokens
• Groq API key → console.groq.com

Backend

# Clone the repo
git clone https://github.com/anshhh1101/studymind.git
cd studymind/backend

# Install dependencies
pip install -r requirements.txt

# Create .env file
echo "HF_KEY=your_huggingface_token" >> .env
echo "GROQ_KEY=your_groq_key" >> .env
echo "DATABASE_URL=postgresql://user:password@localhost/studymind" >> .env

# Run the server
uvicorn main:app --reload --port 8000

Backend runs at: http://localhost:8000 API docs at: http://localhost:8000/docs

Frontend

cd studymind/frontend

# Install dependencies
npm install

# Point to local backend (edit App.jsx line 3)
# const API = "http://localhost:8000";

# Start dev server
npm run dev

Frontend runs at: http://localhost:5173

---

🌍 Deployment

Service	Platform	URL
Frontend	Vercel	studymindv2-xi.vercel.app
Backend	Render	studymind-jqmn.onrender.com
Database	Render PostgreSQL	Internal (Singapore region)

Environment Variables (Render)

Key	Description
HF_KEY	HuggingFace API token (hf_...)
GROQ_KEY	Groq API key (gsk_...)
DATABASE_URL	Render internal PostgreSQL connection string

---

📡 API Reference

GET /

Health check

{ "status": "StudyMind API is running" }

POST /upload

Upload a PDF file

• Body: multipart/form-data with file field
• Response:

{ "message": "PDF uploaded successfully. 6 chunks stored." }

POST /ask

Ask a question about the uploaded PDF

• Body:

{ "question": "What are the key features of the dashboard?" }

• Response:

{ "answer": "The dashboard includes KPI cards, a sentiment slicer, a Bing Map..." }

---

⚠️ Known Limitations

• Single document — uploading a new PDF replaces the previous one
• No conversation memory — each question is independent
• Free tier cold starts — first request after inactivity may take ~50 seconds
• No authentication — app is publicly accessible

---

🔮 Future Improvements

• Top-k retrieval — use 3 best chunks instead of 1 for richer context
• Streaming responses — stream LLM output token by token
• Multi-document support — query across multiple uploaded PDFs
• Conversation history — maintain context across questions
• pgvector — replace JSONB with native vector similarity for scale
• User authentication — secure per-user document storage
• Mobile responsive improvements

---

👨‍💻 Author

Anshuman Dev

• GitHub: @anshhh1101
• B.Tech | Roll No: 23051168

---

Built with ❤️ using RAG Pipeline · FastAPI · HuggingFace · Groq · PostgreSQL · React

⭐ Star this repo if you found it useful!