使用PHP和Claude大模型构建智能文档信息提取系统

Project Overview

This project introduces an intelligent document information extraction system built using PHP and Anthropic's Claude large language model. It supports extracting structured data from PDF files and images (JPEG, PNG, WebP, GIF) for various document types such as ID cards, passports, insurance policies, etc.

Source Info:

• Author/Maintainer: vikashlohiya
• Platform: GitHub
• Repo Link: Extract-text-from-images-using-AI
• Update Time: 2026-06-08T02:18:34Z

Background & Core Goals

In the digital transformation era, manual data entry from paper docs/scans is inefficient and error-prone. With the rise of LLM and multi-modal AI, this project aims to automate the process.

Core objectives:

1. Enable developers to build solutions with minimal code to extract structured data from diverse documents.
2. Return standardized JSON format data for docs like Aadhar (India ID), PAN (India tax ID), passports, insurance policies, birth certificates.

Technical Architecture & Principles

The system centers around Claude API's visual understanding capabilities, with key components:

1. File Upload & Preprocessing: Accepts PDF (≤32MB) and images (≤20MB), checks type/size to meet API limits.
2. Base64 Encoding & API Request: Converts file content to Base64; uses document type for PDFs and image type for images in requests.
3. Structured Output via JSON Schema: Predefines schemas for different docs (e.g., Aadhar includes card number, name, DOB; passport includes passport number, validity, etc.) to ensure Claude returns expected JSON format.

Code Implementation Details

The core function extractDocumentData handles the full flow:

1. Validation: Checks file type and size.
2. Schema Selection: Chooses the appropriate JSON Schema based on document type.
3. API Call: Uses PHP's cURL library to communicate with Claude API (model: claude-sonnet-4-6, timeout:30s).
4. Response Handling: Checks for API errors, extracts JSON results, and adds metadata (doc type, file type, MIME type, extraction timestamp).

Application Scenarios & Value

1. Enterprise Document Automation: Reduces manual entry for insurance companies, banks, HR departments.
2. KYC & Identity Verification: Accelerates processes in finance by extracting info from ID cards/passports.
3. Archive Digitization: Complements OCR for complex, non-fixed format docs.

Improvement Directions

1. Error Handling: Add retry mechanisms for network timeouts and API rate limits.
2. Doc Type Expansion: Support more types (invoices, contracts, medical reports) via dynamic schema loading.
3. Async Processing: Use message queues (RabbitMQ/Redis) for batch document handling.
4. Data Validation: Add checks for data合理性 (e.g., date format) and confidence scoring for low-confidence results.

Summary & Reflections

This project combines multi-modal LLM capabilities with PHP to solve real-world document processing problems. It lowers the barrier to AI application—developers don’t need deep ML knowledge or custom model training to build production-ready systems. It’s a valuable reference for teams with limited resources looking to explore AI solutions.