Semantic search in documentation to enhance the efficiency of UI component generation using language models

Abstract

Large software projects produce extensive documentation, which developers need to access efficiently. This paper presents a lightweight pipeline that converts documentation into a searchable knowledge base for generative LLMs in automated UI code generation. The system combines document crawling, chunking, vector embeddings, and semantic search. Evaluation on representative test cases shows that Retrieval-Augmented Generation (RAG) improves code accuracy and framework compliance compared to baseline LLM generation.