{"repo":"abdo-Mansour/axetract","free":true,"listed":false,"github":"https://github.com/abdo-Mansour/axetract","clone":"git clone https://github.com/abdo-Mansour/axetract.git","description":"Low-Cost Cross-Domain Web Structured Information Extraction using specialized LoRA adapters.","language":"Python","stars":21,"topics":["ai-agents","cross-domain","data-mining","html-parsing","information-extraction","llm","lora","nlp","qwen","structured-data-extraction"],"license":"MIT","category":"ai-agents","readme_excerpt":"AXEtract Low-Cost Cross-Domain Web Structured Information Extraction AXEtract is a high-performance, low-cost framework for extracting structured data from web pages. Based on the paper \"AXE: Low-Cost Cross-Domain Web Structured Information Extraction\" , it optimizes the extraction pipeline by using specialized LoRA adapters for pruning and query-specific extraction, enabling state-of-the-art results with small models (e.g., Qwen3-0.6B). 🚀 Key Features - 🎯 Specialized LoRA Adapters : Uses task-specific adapters for DOM pruning and structured extraction, achieving high accuracy with minimal token overhead. - ✂️ Smart DOM Pruning : Classifies and prunes irrelevant HTML nodes before passing them to the extractor, significantly reducing context window usage and costs. - 📍 Grounded XPath Resolution (GXR) : Automatically maps extracted JSON fields back to their original source XPaths in the DOM for verification and grounding. - ⚡ High-Throughput Pipeline : Built-in support for multiple LLM engines, including vLLM for production-grade serving and HuggingFace for local research. - 🌐 Cross-Domain Versatility : Designed to generalize across various web domains (e-commerce, real estate, listings) without needing domain-specific rules. 🛠️ Architecture AXEtract follows a three-part decoupled pipeline for maximum efficiency: 1. Preprocessor : Fetches raw HTML and chunks it into manageable, token-aware fragments. 2. AI Extractor : Divided into two stages: - Pruner : A lightweight LLM (","default_branch":null,"files":null,"tree":[],"storefront":"/r/abdo-Mansour","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/abdo-Mansour/axetract/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}