主 题：[Open Source] Self-Healing Selector Engine for Web Scraper, Passed Third-Party Audit, Seeking Collaborators for Tauri / LLM Fine-Tuning / SaaS
发 布 者：Uly
标签分类： 技术
时 间：2026-08-03 17:38:47
内容预览：I open-sourced a YAML-driven web scraper that heals broken selectors with an LLM. Then I paid an AI auditor to rip through the entire codebase. Here is what it found. 🧵1/ What this thing doesWrite a YAML config → launch a real Chrome browser (CDP) → extract fields you want → JSON / JSONL / CSV.Page changed? Broken selector? It auto-repairs with an LLM—but only when the first two layers fail. You pay zero API cost 99% of the time.2/ The architectureLayer 1: plain CSS selectors (fast, free)Layer 2: Scrapling adaptive recovery (similarity matching)Layer 3: LLM repair (Gemini / Qwen, calls only on real failure)Config-driven. No hard-coded site logic. One YAML = one site. Works on any public page you are authorized to access.3/ Why I built thisI participated in Alibaba's anti-bot drills. My scr
直达链接： https://www.nodeseek.com/post-855458-1
发 布 者：Uly
标签分类： 技术
时 间：2026-08-03 17:38:47
内容预览：I open-sourced a YAML-driven web scraper that heals broken selectors with an LLM. Then I paid an AI auditor to rip through the entire codebase. Here is what it found. 🧵1/ What this thing doesWrite a YAML config → launch a real Chrome browser (CDP) → extract fields you want → JSON / JSONL / CSV.Page changed? Broken selector? It auto-repairs with an LLM—but only when the first two layers fail. You pay zero API cost 99% of the time.2/ The architectureLayer 1: plain CSS selectors (fast, free)Layer 2: Scrapling adaptive recovery (similarity matching)Layer 3: LLM repair (Gemini / Qwen, calls only on real failure)Config-driven. No hard-coded site logic. One YAML = one site. Works on any public page you are authorized to access.3/ Why I built thisI participated in Alibaba's anti-bot drills. My scr
直达链接： https://www.nodeseek.com/post-855458-1
