Research Paper RAG System

A powerful AI-powered system for ingesting ML/AI research papers and answering deep technical questions with accurate citations.

Features

• PDF Ingestion: Extract text, equations, and structure from research papers
• Semantic Search: Find relevant content using vector embeddings
• Technical Q&A: Answer complex questions about paper content
• Math Explanation: Break down equations into plain English
• Paper Comparison: Compare methodologies across multiple papers
• Exact Citations: All answers include specific page/section references
• Multiple Interfaces: CLI, REST API, and interactive chat

Example Queries

"Explain PPO like I'm new to RL"
"How is DQN different from PPO?"
"What are the limitations of this paper?"
"Summarize the method section in simple terms"
"What does this loss function mean?"

Quick Start

1. Clone and Setup

cd D:\RAG

# Create virtual environment
python -m venv .venv

# Activate (Windows)
.venv\Scripts\activate

# Activate (macOS/Linux)
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

2. Configure API Keys

Create a .env file in the project root:

# Required: At least one LLM provider
OPENAI_API_KEY=sk-your-openai-api-key-here
ANTHROPIC_API_KEY=sk-ant-your-anthropic-api-key-here

# Optional: Customize settings
LLM_PROVIDER=openai
LLM_MODEL=gpt-4-turbo-preview
EMBEDDING_MODEL=text-embedding-3-small

3. Ingest Papers

# Single paper
python main.py ingest path/to/paper.pdf

# Directory of papers
python main.py ingest ./papers/ --recursive

4. Ask Questions

# Command line
python main.py ask "Explain the attention mechanism"

# Interactive mode
python main.py interactive

# Start API server
python main.py serve

Project Structure

D:\RAG/
├── main.py                 # CLI entry point
├── requirements.txt        # Python dependencies
├── pyproject.toml         # Project configuration
├── .env                   # API keys (create from .env.example)
├── .gitignore
│
├── src/
│   ├── __init__.py
│   ├── config.py          # Configuration management
│   │
│   ├── ingestion/         # PDF processing
│   │   ├── __init__.py
│   │   ├── pdf_parser.py  # Extract text from PDFs
│   │   ├── chunker.py     # Split into retrieval chunks
│   │   └── preprocessor.py # Math extraction, text cleanup
│   │
│   ├── retrieval/         # Vector search
│   │   ├── __init__.py
│   │   ├── embeddings.py  # Generate embeddings
│   │   ├── vector_store.py # ChromaDB storage
│   │   └── retriever.py   # High-level retrieval interface
│   │
│   ├── llm/               # Language model integration
│   │   ├── __init__.py
│   │   ├── chat.py        # LLM chat interface
│   │   ├── prompts.py     # Prompt templates
│   │   └── qa_engine.py   # Main Q&A logic
│   │
│   ├── api/               # REST API
│   │   ├── __init__.py
│   │   └── app.py         # FastAPI application
│   │
│   └── cli/               # Command-line interface
│       ├── __init__.py
│       └── main.py        # Typer CLI
│
├── data/
│   ├── papers/            # Store PDFs here
│   ├── processed/         # Processed paper data
│   └── chroma_db/         # Vector store (auto-created)
│
└── tests/
    └── ...

CLI Commands

Command	Description
ingest <path>	Ingest PDF file(s) into the system
ask "<question>"	Ask a question about indexed papers
compare "<paper1>" "<paper2>"	Compare multiple papers
list	List all indexed papers
stats	Show system statistics
delete "<paper>"	Remove a paper from the index
serve	Start the REST API server
interactive	Start interactive chat mode

REST API Endpoints

Start the server with python main.py serve, then:

Method	Endpoint	Description
POST	/papers/upload	Upload and index a PDF
POST	/ask	Ask a question
POST	/compare	Compare papers
GET	/papers	List indexed papers
DELETE	/papers/{title}	Delete a paper
GET	/stats	Get system statistics

Example API Usage

# Upload a paper
curl -X POST "http://localhost:8000/papers/upload" \
  -F "file=@paper.pdf"

# Ask a question
curl -X POST "http://localhost:8000/ask" \
  -H "Content-Type: application/json" \
  -d '{"question": "Explain the main contribution"}'

Configuration Options

Variable	Default	Description
LLM_PROVIDER	openai	LLM provider (openai/anthropic)
LLM_MODEL	gpt-4-turbo-preview	Model to use
EMBEDDING_MODEL	text-embedding-3-small	Embedding model
CHUNK_SIZE	1000	Characters per chunk
CHUNK_OVERLAP	200	Overlap between chunks
MAX_CONTEXT_TOKENS	8000	Max tokens for context

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black src/ tests/

# Lint
ruff check src/ tests/

Architecture

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   PDF       │────▶│  Chunker    │────▶│  Vector     │
│   Parser    │     │             │     │  Store      │
└─────────────┘     └─────────────┘     └─────────────┘
                                              │
                                              ▼
┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   User      │────▶│  QA         │────▶│  Retriever  │
│   Query     │     │  Engine     │     │             │
└─────────────┘     └─────────────┘     └─────────────┘
                          │
                          ▼
                    ┌─────────────┐
                    │   LLM       │
                    │   (GPT-4)   │
                    └─────────────┘
                          │
                          ▼
                    ┌─────────────┐
                    │  Answer +   │
                    │  Citations  │
                    └─────────────┘