🔍 LangExtract - Structured Data Extractor

Extract structured data from unstructured text using Google's LangExtract library powered by Large Language Models (LLMs). Transform messy documents into clean, organized JSON with precise source grounding.

✨ What It Does

This Actor leverages Google's open-source LangExtract library to:

• Extract entities (people, organizations, products, locations, etc.) from any text
• Identify relationships and attributes between entities
• Ground extractions to exact positions in the source text
• Generate visual reports showing extractions in context

🎯 Use Cases

Industry	Application
HR & Recruiting	Extract candidates, skills, and experience from resumes
Legal	Pull parties, dates, and clauses from contracts
Healthcare	Structure patient info from medical records
Finance	Extract entities from earnings reports and filings
Research	Parse academic papers for citations and findings
News & Media	Identify people, places, and events from articles
E-commerce	Extract product specs from descriptions

🚀 Quick Start

Basic Example

Input:

{
  "provider": "Gemini",
  "apiKey": "YOUR_GEMINI_API_KEY",
  "model": "gemini-2.5-flash",
  "schema": "{\"prompt_description\": \"Extract all people and their job titles.\"}",
  "text": "Sarah Chen was appointed CEO of Acme Corp. Dr. James Wilson serves as CTO."
}

Output:

{
  "extraction": {
    "extractions": [
      {"class": "person", "text": "Sarah Chen", "attributes": {"role": "CEO", "organization": "Acme Corp"}},
      {"class": "person", "text": "Dr. James Wilson", "attributes": {"role": "CTO"}}
    ]
  }
}

🤖 Supported Providers & Models

Google Gemini (Recommended)

Model	Best For
gemini-2.5-flash	Fast, affordable, great quality
gemini-2.0-flash	Balanced performance
gemini-1.5-pro	Complex extractions

OpenAI

Model	Best For
gpt-4o	Highest quality
gpt-4o-mini	Fast and affordable
gpt-4-turbo	Long documents

Anthropic

Model	Best For
claude-3-5-sonnet-latest	Best overall
claude-3-haiku-latest	Fast responses

Groq (Fast Inference)

Model	Best For
llama-3.3-70b-versatile	High quality
llama-3.1-8b-instant	Ultra-fast
mixtral-8x7b-32768	Long context
gemma2-9b-it	Efficient

OpenAI-Compatible (Custom/Local)

Use any OpenAI-compatible API by providing a custom Base URL:

• Ollama: http://localhost:11434/v1
• Azure OpenAI: https://YOUR_RESOURCE.openai.azure.com
• Together AI, Anyscale, etc.

📥 Input Parameters

Parameter	Type	Required	Description
provider	enum	✅	LLM provider: Gemini, OpenAI, Anthropic, Groq, OpenAI-Compatible
apiKey	secret	✅	Your API key for the selected provider
model	string	✅	Model name (e.g., gemini-2.5-flash)
schema	json	✅	Extraction instructions (see Schema section)
text	string	⚪	Raw text to process
urls	array	⚪	URLs to fetch and process
baseUrl	string	⚪	Custom API endpoint (for OpenAI-Compatible)
systemPrompt	string	⚪	Custom persona (e.g., "Act as a legal expert")
batchMode	boolean	⚪	Enable parallel processing (default: true)
debug	boolean	⚪	Save debug info to Key-Value Store
trackTokens	boolean	⚪	Report token usage (default: true)

📝 Schema Format

Simple Format

Just describe what you want to extract:

{
  "prompt_description": "Extract all people mentioned with their job titles and organizations."
}

Advanced Format (with examples)

Provide examples for better accuracy:

{
  "prompt_description": "Extract people and their roles. Use exact text from the source.",
  "examples": [
    {
      "text": "John Smith is the CEO of TechCorp.",
      "extractions": [
        {
          "extraction_class": "person",
          "extraction_text": "John Smith",
          "attributes": {"role": "CEO", "organization": "TechCorp"}
        }
      ]
    }
  ]
}

📤 Output

Dataset

Each extraction is pushed to the Apify Dataset:

{
  "source_index": 0,
  "extraction": {
    "extractions": [
      {"class": "person", "text": "...", "attributes": {...}, "position": {"start": 0, "end": 10}}
    ],
    "source_text": "..."
  },
  "status": "success"
}

Key-Value Store

• EXTRACTION_REPORT.html - Interactive visual report
• TOKEN_USAGE - Token count and cost estimate (if enabled)
• DEBUG_* - Debug info (if enabled)

🔒 Security

• API keys are encrypted and never logged
• Data is processed transiently - not stored beyond your run
• Runs in sandboxed containers on Apify infrastructure

💡 Tips

1. Better results: Provide examples in your schema for complex extractions
2. Faster runs: Use gemini-2.5-flash or llama-3.1-8b-instant for speed
3. Long documents: The Actor automatically chunks and processes large texts
4. Multiple sources: Add multiple URLs or texts for batch processing
5. Custom personas: Use systemPrompt to guide extraction style

🔗 Links

• LangExtract GitHub
• Get Gemini API Key
• Get OpenAI API Key

📜 License

Apache 2.0 - Built with Google's LangExtract library.