LLM-Powered Self-Healing for Fragile Web Scrapers

Web scrapers are fragile: a single website redesign breaks CSS selectors and halts data pipelines, requiring manual investigation and deployment cycles. This system demonstrates how local LLMs can automate selector repair, reducing operational friction from hours to minutes. By treating LLMs as proposers rather than decision-makers and validating all suggestions against live HTML, it achieves reliable automation without sacrificing safety.