📄 PDF to Knowledge Graph Extractor

Free LLM-powered information extraction from academic papers using Google Gemini

🙏 Acknowledgements & Inspiration

This project is inspired by and builds upon the excellent work in the LLM and Knowledge Graph community.

Original Reference & Evolution

This project evolved from the Structured RAG implementation by thu-vu92. The original repository provided a foundational approach to:

• PDF document loading and chunking with LangChain
• Vector database creation using Chroma
• Basic RAG pipeline implementation

Original Source: thu-vu92/structured-rag-pdf/data_extraction_llms.ipynb

Key Differences & Extensions in This Project

Aspect	Original (thu-vu92)	This Project
LLM Provider	OpenAI GPT-4o-mini (paid)	Google Gemini (free)
Embeddings	OpenAI text-embedding-ada-002	Google gemini-embedding-001
Output Format	Basic text responses	Structured Pydantic models
Knowledge Graph	Not included	Subject-Predicate-Object triplet extraction
Cost	Requires API credits	100% free

What I Added / Changed

1. Switched to Google Gemini - Making it accessible with free tier
2. Triplet extraction - Converting text to knowledge graph format

🎯 What does this project do?

This notebook extracts structured knowledge from PDF documents using:

• Google Gemini (free tier!) for LLM-powered extraction
• RAG (Retrieval-Augmented Generation) for semantic search
• Structured output with Pydantic models
• Knowledge Graph triplets (Subject-Predicate-Object)

✨ Features

Feature	Description
📥 PDF Loading	Load any academic PDF
🔍 Semantic Search	ChromaDB with Gemini embeddings
🧠 Structured Extraction	Title, authors, summary, year
🔗 Triplet Extraction	Subject → Predicate → Object
💰 100% Free	Uses Google Gemini (no OpenAI costs)

🛠️ Tech Stack

• LLM: Google Gemini (gemini-flash-latest)
• Embeddings: gemini-embedding-001
• Vector DB: Chroma
• Framework: LangChain
• Output: Triplet Output in Console

🚀 Quick Start

1. Get a Google Gemini API Key

• Go to Google AI Studio
• Click "Get API Key"
• Copy your key

2. Setup

Open .ipynb file in Colab

Set up environment variables

Create file called ".env" at project directory and add your GOOGLE_API_KEY=[YOUR_KEY]

Switching Between LLMs

# For Open AI (pay to use)
llm = ChatOpenAI(model="gpt-4o-mini", api_key=OPENAI_API_KEY)

# For Gemini (free with limited request rate)
llm = ChatGoogleGenerativeAI(model="gemini-flash-latest", api_key=GOOGLE_API_KEY)