Promote Sign in Sign up

Repository

LangExtract-RAG

LangExtract-Enhanced RAG System

PromtEngineer
Repository

README

LangExtract-Enhanced RAG System

🚀 Overview

Here is a video to get you started

This project demonstrates how Google's LangExtract transforms traditional RAG (Retrieval-Augmented Generation) systems by adding intelligent metadata extraction.

The Problem with Traditional RAG

Traditional RAG systems treat all documents equally, leading to:

  • Mixed contexts from different versions
  • Conflicting information in responses
  • Generic "it depends" answers
  • No version or service awareness

The LangExtract Solution

LangExtract adds a metadata intelligence layer that:

  • Extracts structured metadata (versions, services, document types)
  • Filters documents BEFORE retrieval
  • Provides precise, version-specific answers
  • Eliminates context confusion

🏗️ Architecture

Traditional RAG Flow

Documents → Chunks → Embeddings → Vector DB → Search ALL

LangExtract-Enhanced RAG Flow

Documents → LangExtract → Metadata + Chunks → Vector DB → Filter → Search Subset

🔧 Installation

Prerequisites

  • Python 3.8+
  • Google Cloud API key (for LangExtract/Gemini)
  • OpenAI API key (optional, for embeddings)

Setup

  1. Clone the repository

    git clone https://github.com/PromtEngineer/langextract_rag.git
cd langextract_rag

  2. Install dependencies

    pip install -r requirements.txt

  3. Set up environment variables
    Create a .env file:

    GOOGLE_API_KEY="your-google-api-key"
OPENAI_API_KEY="your-openai-api-key"  # Optional

🎯 Quick Start

Run the Demo

# Run the working demo with real LangExtract
python langextract_rag.py

Query: "How do I authenticate with OAuth in version 2.0?"

Traditional RAG Response:

"The platform supports multiple authentication methods:
- Version 2.0 uses OAuth 2.0...
- Version 1.0 uses API keys (deprecated)...
Note: Different versions handle auth differently."

LangExtract RAG Response:

"To authenticate with OAuth in version 2.0:
1. Send POST to https://api.platform.com/auth/oauth2/token
2. Include client_id, client_secret, grant_type, scope
3. Receive access token (1 hour) and refresh token (30 days)"

🎓 How It Works

1. Metadata Extraction

LangExtract analyzes documents and extracts:

  • Service names (e.g., "Authentication API", "Storage Service")
  • Version numbers (e.g., "2.0", "1.0")
  • Document types (reference, guide, troubleshooting)
  • Key features (rate limits, deprecation notices)

2. Smart Filtering

Query parser extracts filters from natural language:

  • "version 2.0" → filter: {version: "2.0"}
  • "authentication" → filter: {service: "Authentication API"}
  • "troubleshoot" → filter: {doc_type: "troubleshooting"}

3. Precise Retrieval

Instead of searching all documents:

  • Apply metadata filters first
  • Search only within relevant subset
  • Return focused, accurate results

🙏 Acknowledgments