LLM Extract

Extract structured data from documents using LLMs. Inspired by LangExtract.

Key Features:

• Multi-worker parallel processing for large documents
• Document processing with PDF text extraction and OCR fallback
• Structured data extraction using LLMs with few-shot learning

Installation

npm install llm-extract

Supported Providers

| Provider | Status | |----------|--------| | OpenAI | ✅ Available | | Azure OpenAI | ✅ Available | | Anthropic Claude | 🔄 Coming Soon | | Google Gemini | 🔄 Coming Soon |

Usage

import { LanguageExtractor, OpenAIProvider } from 'llm-extract';

// Setup with OpenAI
const provider = new OpenAIProvider({
  apiKey: "sk-your-openai-api-key",
  model: "gpt-4o"
});

// Or with Azure OpenAI
// const provider = new AzureOpenAIProvider({
//   apiKey: "your-api-key",
//   endpoint: "https://your-endpoint.openai.azure.com/",
//   deploymentName: "gpt-4",
//   model: "gpt-4"
// });

const extractor = new LanguageExtractor();
extractor.setLLMProvider(provider);

// Extract from text
const result = await extractor.extract({
  textOrDocuments: "Contract with John Doe dated 2024-01-15",
  promptDescription: "Extract names and dates",
  temperature: 0.1
});

console.log(result.extractions);
// [{ extraction_class: "name", extraction_text: "John Doe" }, ...]

Document Processing

import { PDFOCRProcessor, ImageOCRProcessor } from 'llm-extract';

// Step 1: Process document to extract text
const pdfProcessor = new PDFOCRProcessor();
const parsedDoc = await pdfProcessor.parseDocument(pdfBuffer, {
  fallbackToBasic: true,
  config: {
    tesseract: { language: 'eng' },
    pdf2pic: { density: 200 }
  }
});

// Step 2: Extract structured data from text
const result = await extractor.extract({
  textOrDocuments: parsedDoc.extractedText,
  promptDescription: "Extract invoice details"
});

Examples with Training Data

const result = await extractor.extract({
  textOrDocuments: "Agreement with ABC Corp on 2024-01-15",
  promptDescription: "Extract companies and dates",
  examples: [
    {
      text: "Contract with XYZ Ltd dated 2023-12-01",
      extractions: [
        { extraction_class: "company", extraction_text: "XYZ Ltd" },
        { extraction_class: "date", extraction_text: "2023-12-01" }
      ]
    }
  ]
});

License

MIT