Intent-Based Software Development Using Large Language Models: Transforming Requirements into Verified Application Components

Abstract

ABSTRACT Intent-Based Software Development (IBSD) represents a paradigm shift in software engineering, transitioning from explicit procedural programming to high-level, declarative intent specification powered by Large Language Models (LLMs). While state-of-the-art LLMs demonstrate unprecedented capabilities in code generation, translating ambiguous, natural language human requirements into robust, functionally correct, and formally verified application components remains a critical research challenge. Non deterministic hallucination, subtle logical defects, and security vulnerabilities inherent in LLM outputs impede their direct integration into mission-critical software ecosystems. This review paper comprehensively examines the emergence of LLM-driven intent-based software engineering, dissecting the architectural pipelines, semantic mapping techniques, formal verification frameworks, and automated feedback loops that facilitate the transformation of raw requirements into executable, verified components. We evaluate current methodologies including Retrieval-Augmented Generation (RAG), Abstract Syntax Tree (AST) validation, Satisfiability Modulo Theories (SMT) solving, and neuro-symbolic feedback systems. Furthermore, we analyze empirical findings across benchmark datasets, identify core technical research gaps, outline domain limitations, and map out promising future research directions. This paper serves as a rigorous scientific foundation for researchers and industrial practitioners advancing the frontier of automated, intent-driven software synthesis.

KEYWORDS: Intent-Based Software Development, Large Language Models, Code Generation, Formal Verification, SMT Solvers, Neuro-Symbolic Systems, Automated Program Synthesis.