WebPilot AI: Autonomous Web Interaction using LLMs

This paper presents WebPilot AI, an autonomous web agent framework designed to enable Large Language Models (LLMs) to interact with real-time web environments. The rapid advancement of Large Language Models (LLMs) has significantly enhanced natural language processing and automated reasoning capabilities. However, their reliance on static training data limits their ability to access real-time information, often leading to hallucinations. This paper presents WebPilot AI, an autonomous web agent framework that enables LLMs to interact with live web environments. The proposed system integrates dynamic headless browser rendering, semantic parsing, and DOM minimization techniques to extract high-quality, token-efficient content from modern web applications. A formal mathematical model is introduced to evaluate token efficiency and noise reduction. Experimental benchmarking demonstrates that WebPilot AI achieves superior retrieval accuracy and token efficiency compared to existing solutions such as Perplexity AI, Bing Browse, and MultiOn. Furthermore, the paper explores the integration of WebPilot AI into cybersecurity pipelines for detecting AI-driven phishing and web-based threats. The results highlight the potential of autonomous web agents in bridging the gap between static LLM knowledge and dynamic real-world data.