Skip to content

README.md

Latest commit

 

History

History
266 lines (200 loc) · 7.49 KB

README.md

File metadata and controls

266 lines (200 loc) · 7.49 KB

🧠 Smart Study Assistant — AI Tutor with Zero-Shot Prompting, RAG & Function Calling

A personalized AI-powered tutor that intelligently adapts to any question using Zero-Shot Prompting, Retrieval-Augmented Generation (RAG), and Function Calling. This assistant helps students study smarter by automatically optimizing AI responses for different subjects and learning contexts.

✨ Built for BTech prep, competitive exams, and lifelong learners with intelligent prompt adaptation.

🚀 Features

Feature Description
🎯 Zero-Shot Prompting Automatically detects question type (math, science, history, etc.) and optimizes AI responses accordingly.
🔍 Question Answering (RAG) Ask questions from uploaded notes or PDFs. Get AI-generated answers grounded in source content.
🧾 Summarization Upload documents and get key-point summaries for revision.
🎓 Flashcard Generator Auto-generate Q&A flashcards for active recall and spaced repetition.
📋 Quiz Generator (MCQs) Convert study material into quizzes with correct options and distractors.
⏰ Reminder Scheduler Schedule study sessions or revision reminders using natural language.
🧠 Adaptive Prompting Subject-specific prompts with difficulty levels (beginner/intermediate/advanced).

🧠 Tech Stack

Layer Tools Used
⚙️ Backend Node.js, Express.js, JavaScript (ES Modules)
📄 Document Parsing JavaScript parsing utilities
🧠 Language Model Google Gemini API (@google/genai)
🔍 Vector DB In-memory context management
🎯 Structured Output JSON schemas for flashcards, summaries, quizzes
🧠 Function Calling Reminder APIs, quiz generation, schedule builder
🌐 Frontend (optional) React.js + Tailwind CSS or Postman/Bruno for API demo
🔧 Prompting Strategies Zero-Shot, One-Shot, Multi-Shot, Chain-of-Thought

🏗️ Architecture

graph TD
    A["User Input (Text or Voice)"] --> B["Prompt + Notes"]
    B --> C["Retrieval Engine (RAG)"]
    C --> D["LLM (GPT-4 with Function Calling)"]
    D --> E{"Structured Output?"}
    E -->|Flashcards| F["Flashcard JSON"]
    E -->|Quiz| G["Quiz JSON"]
    E -->|Answer| H["Text Output"]
    D --> I["Call Reminder API"]
    F & G & H --> J["Frontend / API Response"]
Loading

✨ Sample Use Cases

🗂️ Upload syllabus or PDF notes, then try:

🔹 Question Answering

"Explain Newton's Second Law with examples from my uploaded physics notes."

🔹 Flashcard Generation

"Generate 5 flashcards for the photosynthesis topic in my biology.pdf."

🔹 Quiz Generation

"Create a quiz with 5 MCQs from the chapter on Semiconductor Devices."

🔹 Zero-Shot Prompting (NEW!)

# Automatic subject detection and optimization
"Solve 2x + 5 = 15"  # → Detects math, shows step-by-step solution
"What is photosynthesis?"  # → Detects science, provides detailed explanation with examples
"When did World War I start?"  # → Detects history, gives chronological context

# Manual control with options
node zero-shot.js --type math --level beginner "Explain derivatives"
node zero-shot.js --task tutorial "How to solve quadratic equations"

🔹 Reminder Scheduling (Function Calling)

"Remind me to revise Operating Systems every Monday at 7 PM."

⚙️ How It Works (Behind the Scenes)

  1. Upload Document / Notes
    PDF, Markdown, or plain text is parsed and chunked for embedding.

  2. Vector Store Creation
    Notes are embedded and stored in a vector DB (e.g., Chroma).

  3. RAG Pipeline
    On user query, relevant chunks are retrieved and passed into the LLM.

  4. Prompting + Role Instructions
    Prompts are engineered to elicit tutor-like answers, summaries, or flashcards.

  5. Structured Output
    JSON schema (with tool calling or OpenAI function calling) enables:

    • Reminders (via API)
    • Flashcard/Quiz output
    • Visual rendering on frontend

🛠️ Setup Instructions

1. Clone the repo

git clone https://github.com/kalviumcommunity/Smart-Study-Assistant.git
cd smart-study-assistant

2. Install dependencies

npm install
# or if using Python-based document processing
pip install -r requirements.txt

3. Set up environment variables

Create .env and add:

4. Run backend server

npm run dev

5. Upload PDFs & Test

Use Postman or frontend UI to upload a file and ask questions.

🧪 API Examples

POST /ask-question

{
  "query": "What is Ohm's Law?",
  "documentId": "physics-notes"
}

POST /generate-flashcards

{
  "topic": "Photosynthesis",
  "documentId": "bio.pdf"
}

POST /schedule-reminder

{
  "reminder": "Revise DSA on weekends",
  "time": "Saturday 6 PM"
}

🚀 Quick Start - Zero-Shot Prompting

Command Line Usage

# Navigate to backend directory
cd backend

# Basic usage (automatic detection)
node zero-shot.js "What is photosynthesis?"
node zero-shot.js "Solve 3x + 7 = 22"

# With specific options
node zero-shot.js --type math --level beginner "Explain calculus"
node zero-shot.js --task tutorial "How to write Python functions"

# See all available options
node zero-shot.js --types
node zero-shot.js --help

# Run demonstration
node demo-zero-shot.js

API Usage

# Standard chat with zero-shot prompting
curl -X POST http://localhost:3000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Solve 2x + 5 = 15", "options": {"promptType": "math"}}'

# Advanced zero-shot endpoint
curl -X POST http://localhost:3000/chat/zero-shot \
  -H "Content-Type: application/json" \
  -d '{"message": "Explain machine learning", "promptType": "programming", "level": "intermediate"}'

# Get available prompt types
curl http://localhost:3000/chat/types

🧰 Prompt Engineering Examples

System Prompt:

You are a friendly and helpful tutor for engineering students. You answer using the uploaded study notes only. Keep explanations concise and exam-focused.

User Prompt:

Based on my Operating Systems notes, explain multithreading with examples.

Output Format (Structured):

{
  "topic": "Multithreading",
  "explanation": "Multithreading allows multiple threads to run concurrently...",
  "examples": ["Thread A for UI", "Thread B for I/O operations"]
}

🧠 RAG Setup

  • Load document using LangChain / LlamaIndex
  • Split into chunks (~500 tokens)
  • Embed using OpenAI Embeddings
  • Store in FAISS/Chroma vector store
  • Query = Embed + Similarity search + Prompt injection

💡 Future Features

  • 🗣️ Voice Interaction via Whisper + TTS
  • 📚 Notion/Obsidian integration
  • 📱 Mobile version in React Native
  • 📊 Analytics Dashboard for user performance
  • 🎮 Gamification with rewards & leaderboards

👨‍💻 Author

Chethan Regala
Second year BTech Student(Kalvium) | Passionate about Generative AI, Full Stack, and Building Scalable Products
📧 chethanregala123@gmail.com
🔗 LinkedIn | GitHub