AI-Driven Full–stack Software Development: Transforming Designs and Natural Language into High-Quality Applications

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Devally is an AI-driven full-stack software development system that translates design assets and natural language requirements into functional front-end and back-end code. It supports multiple input modalities for frontend generation, including screenshots, figma-like designs, templates, and hand-drawn sketches, while backend functionality is derived from structured human-readable requirements. The system architecture consists of four core modules: (i) a code repository that can be enabled or disabled based on desired performance trade-offs and containing reusable code snippets and templates ; (ii) a requirements analyzer that classifies and processes inputs while leveraging stored code assets; (iii) a large language model (LLM) augmented with vision transformers to enhance contextual understanding during code generation; and (iv) an integration module that composes, adapts, and merges retrieved or newly generated code into fully functional applications. Devally automates key development tasks, optimizing time, cost, and effort while improving code quality, scalability, and maintainability. Empirical evaluations across multiple case studies demonstrate that Devally consistently produces production-ready applications, bridging the gap between design concepts and their technical implementation. With the code repository module disabled, Devally's performance remains comparable to state-of-the-art LLMs. When enabled, however, it significantly outperforms all other evaluated LLM-based approaches, demonstrating superior accuracy, efficiency, and adaptability in a variety of software development scenarios.

Article activity feed