Browser Automation using LLM

This paper presents a modular framework for browser automation using large language models (LLMs). The system converts raw HTML into a token-efficient JSON representation of the DOM, reducing token usage by over 90% while retaining essential structural and semantic information. By combining structured prompting, error recovery mechanisms, and Selenium-based execution, the framework reliably translates natural language instructions into browser operations across diverse websites. Evaluations on hundreds of user commands demonstrate high success rates, with most tasks completing in real time. This approach highlights the potential of LLM-driven agents to perform complex web interactions efficiently, offering a reusable skeleton for developers to adapt in building AI-powered automation solutions.