An LLM-Based Full-Stack Code Generation and Validation Framework

Author

Gourav, Arjoo

Abstract

 Large Language Models (LLMs) have demonstrated strong capability in translating natural-language requirements into working code, yet most existing systems operate on a single layer of an application generating either frontend components or backend services in isolation. This paper proposes an LLM-based full-stack code generation and validation framework that converts a single natural-language requirement into coordinated frontend components (React), backend REST APIs (Node.js), and database schemas and operations (PostgreSQL), and then validates the resulting application as one coherent system rather than as disconnected parts. The central research problem addressed is that independently generated frontend and backend code may each be individually functional yet incompatible when integrated, owing to mismatched data contracts, inconsistent API schemas, and divergent assumptions about state and validation. The proposed framework introduces a cross-layer integration-contract engine that binds the three generated layers to a shared specification, followed by an automated validation pipeline comprising unit testing, integration testing, end-to-end testing, and static security analysis. The framework is evaluated using three criteria: integration success rate, functional correctness, and security defect density. Preliminary evaluation design indicates that explicit contract-level validation substantially reduces integration failures compared to independent, layer-isolated generation. The paper contributes a reusable architecture, an evaluation methodology, and a research agenda for reliable, validated full-stack code generation using LLMs.

Keywords

Large Language Models, Full-Stack Code Generation, Software Integration, Automated Validation, React, Node.js, PostgreSQL, Security Analysis

Full Text:

Download Paper PDF

References

[1] Chen, M., Tworek, J., Jun, H., et al. (2021). Evaluating large language models trained on code. arXiv preprint.

[2] Austin, J., Odena, A., Nye, M., et al. (2021). Program synthesis with large language models. arXiv preprint.

[3] Nijkamp, E., Pang, B., Hayashi, H., et al. (2023). CodeGen: An open large language model for code with multi-turn program synthesis. International Conference on Learning Representations.

[4] Qian, C., Cong, X., Yang, C., et al. (2023). Communicative agents for software development. arXiv preprint.

[5] Hong, S., Zhuge, M., Chen, J., et al. (2023). MetaGPT: Meta programming for a multi-agent collaborative framework. arXiv preprint.

[6] Pearce, H., Ahmad, B., Tan, B., Dolan-Gavitt, B., & Karri, R. (2022). Asleep at the keyboard? Assessing the security of GitHub Copilot's code contributions. IEEE Symposium on Security and Privacy.

[7] Fried, D., Aghajanyan, A., Lin, J., et al. (2023). InCoder: A generative model for code infilling and synthesis. International Conference on Learning Representations.

[8] Li, Y., Choi, D., Chung, J., et al. (2022). Competition-level code generation with AlphaCode. Science, 378(6624), 1092–1097.

[9] Ross, S. I., Martinez, F., Houde, S., Muller, M., & Weisz, J. D. (2023). The programmer's assistant: Conversational interaction with a large language model for software development. Proceedings of the 28th International Conference on Intelligent User Interfaces.

[10] Fowler, M. (2018). Patterns of Enterprise Application Architecture (2nd ed.). Addison-Wesley.

Share your valuable work from Social Media Buttons