AI-Powered Study Buddy for MISM Students

An intelligent study assistant that helps MISM students at CMU learn from their course materials through AI-powered summaries, practice questions, and interactive Q&A using Retrieval-Augmented Generation (RAG).

Features

Core Functionality

• Document Processing: Upload and process PDFs, PowerPoint slides, Word documents, and text files
• RAG-Powered Q&A: Ask questions and get accurate answers grounded in your course materials
• Smart Summaries: Generate concise summaries of topics or entire documents
• Practice Questions: Auto-generate diverse question types:
  • Multiple Choice Questions (MCQs)
  • True/False
  • Fill-in-the-Blanks
  • Short Answer
  • Match-the-Following
  • Long Answer/Essay Questions
• Evaluation Framework: Built-in RAGAS and DeepEval metrics for quality assessment
• User Feedback: Collect and analyze user feedback for continuous improvement

Architecture

AI-study-buddy/
├── backend/              # FastAPI backend server
│   ├── main.py          # API endpoints
│   ├── config.py        # Configuration management
│   └── requirements.txt
├── frontend/            # Streamlit UI
│   ├── app.py          # Main application
│   └── requirements.txt
├── utils/              # Core utilities
│   ├── document_processor.py  # Document parsing and chunking
│   ├── vector_store.py        # Vector database and RAG pipeline
│   └── content_generator.py  # LLM-powered content generation
├── evaluation/         # Evaluation framework
│   └── evaluator.py   # RAGAS and DeepEval integration
├── data/              # Data storage
│   ├── raw/          # Uploaded documents
│   ├── processed/    # Processed documents
│   └── chromadb/     # Vector database
├── models/           # Model configurations
└── tests/           # Test suite for debugging

Quick Start

Prerequisites

• Python 3.9 or higher
• OpenAI API key (required)
• Git

Setup

1. Get your OpenAI API key: Visit OpenAI API Keys

Installation

1. Set up environment variables

cp .env.example .env
# Edit .env and add your OpenAI API key

2. Install backend dependencies

cd backend
pip install -r requirements.txt

3. Install frontend dependencies

cd ../frontend
pip install -r requirements.txt

Running the Application

1. Start the backend server (Terminal 1)

cd backend
python main.py

The API will be available at http://localhost:8000

2. Start the frontend (Terminal 2)

cd frontend
streamlit run app.py

The UI will open in your browser at http://localhost:8501

Usage Guide

1. Upload Documents

• Navigate to the "Upload Documents" page
• Select a PDF, PPTX, DOCX, or TXT file
• Click "Process Document"
• The system will extract text, create chunks, and generate embeddings

2. Ask Questions

• Go to the "Ask Questions" page
• Type your question about the course materials
• Get AI-generated answers with supporting context
• Rate the answer to help improve the system

3. Generate Summaries

• Visit the "Generate Summary" page
• Choose topic-based or custom query summary
• Receive a concise summary of the content
• Download the summary for later reference

4. Practice Questions

• Select "Practice Questions"
• Choose question type (MCQ, True/False, etc.)
• Specify number of questions
• Review questions and answers for self-assessment

Configuration

Edit .env file to customize:

# OpenAI Settings
OPENAI_API_KEY=your_key_here
OPENAI_MODEL=gpt-4-turbo-preview
OPENAI_EMBEDDING_MODEL=text-embedding-3-small

# RAG Settings
CHUNK_SIZE=1000
CHUNK_OVERLAP=200
TOP_K_RETRIEVAL=5
TEMPERATURE=0.4

# Vector Database
VECTOR_DB_TYPE=chromadb
CHROMADB_PATH=./data/chromadb

RAG Pipeline

The system implements a sophisticated RAG pipeline:

1. Document Processing

   • Extract text from various file formats
   • Split into manageable chunks with overlap
   • Preserve metadata and context

2. Embedding Generation

   • Generate vector embeddings using OpenAI
   • Store in ChromaDB for efficient retrieval
   • Support batch processing for large documents

3. Retrieval

   • Convert queries to embeddings
   • Perform similarity search
   • Retrieve top k most relevant chunks

4. Generation

   • Provide context to LLM
   • Generate accurate, grounded responses
   • Include source references

Evaluation

The system includes comprehensive evaluation:

RAGAS Metrics

• Faithfulness: How factually accurate are the answers?
• Answer Relevancy: How relevant is the answer to the query?
• Context Precision: How precise is the retrieved context?
• Context Recall: How complete is the retrieved context?

DeepEval Metrics

• Answer Relevancy: Semantic relevance of responses
• Faithfulness: Consistency with source material
• Coherence: Logical flow and clarity

User Feedback

• Star ratings (1-5)
• Qualitative comments
• Usage analytics

Security & Privacy

• Documents are stored locally
• No data sharing with third parties
• API keys stored securely in environment variables
• User sessions isolated
• Uploaded files can be deleted anytime

API Documentation

Once the backend is running, visit:

• API Docs: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

Key Endpoints

• POST /upload - Upload and process a document
• POST /query - Ask questions about materials
• POST /summary - Generate summaries
• POST /questions - Generate practice questions
• GET /stats - Get knowledge base statistics

Tips for Best Results

1. Upload Quality Materials: Clear, well-formatted documents work best
2. Specific Questions: More specific queries yield better answers
3. Chunk Size: Adjust based on your document structure
4. Regular Updates: Keep adding new materials for better coverage
5. Provide Feedback: Help improve the system through ratings

---

Built with ❤️ for MISM students at Carnegie Mellon University