Vedant Gudpale,
Omkar Jadhav,
Sujal Shingrut,
Atharva Kumbhar,
Ambuj Kumar,
- Student, Department of Computer Engineering Mumbai University, Vishwaniketan’s Institute of Management, Entrepreneurship & Engineering Technology (iMEET), Maharashtra, India
- Student, Department of Computer Engineering Mumbai University, Vishwaniketan’s Institute of Management, Entrepreneurship & Engineering Technology (iMEET), Maharashtra, India
- Student, Department of Computer Engineering Mumbai University, Vishwaniketan’s Institute of Management, Entrepreneurship & Engineering Technology (iMEET), Maharashtra, India
- Student, Department of Computer Engineering Mumbai University, Vishwaniketan’s Institute of Management, Entrepreneurship & Engineering Technology (iMEET), Maharashtra, India
- Professor, Department of Computer Engineering Mumbai University, Vishwaniketan’s Institute of Management, Entrepreneurship & Engineering Technology (iMEET), Maharashtra,
Abstract
The proliferation of large language models (LLMs) has fundamentally altered the landscape of automated software engineering. However, a significant “deployment chasm” persists between the generation of code artifacts and their realization as production-grade cloud applications. Traditional workflows are marred by the “DevOps Tax”: the high operational overhead of environment configuration, dependency resolution, and infrastructure provisioning. This paper presents a novel, technically adept architecture that bridges this gap through a unified Prompt-to-Production (P2P) pipeline. Our system integrates the llama-3.3-70b-versatile model via Groq-accelerated inference with a client-side WebAssembly (Wasm) execution environment (WebContainers) and an automated AWS S3 static hosting conduit. We provide a mathematical formalization of the code synthesis process and conduct a rigorous empirical evaluation against three benchmark tasks. Results demonstrate a 95% reduction in time-to-deployment (TTD) compared to manual baselines while maintaining architectural integrity and security via Cross-Origin Isolation (COEP/COOP) and Identity and Access Management (IAM) least-privilege principles. We further establish the platform’s novelty by contrasting it with recent 2021-2025 literature, identifying our specific contribution as the first zero-configuration, browser-native deployment state machine. Beyond mere efficiency gains, our architecture addresses the cognitive load inherent in full-stack orchestration by abstracting the underlying virtualization layer. By leveraging the low-latency inference of the Groq Language Processing Unit (LPU)™ alongside a sandboxed Node.js runtime operating entirely within the browser’s main thread, the system eliminates the need for remote build servers. This shift toward “edge-side development” ensures that code is validated in a deterministic environment before being pushed to a global content delivery network (CDN). Consequently, this research provides a scalable blueprint for the next generation of autonomous coding agents, enabling non-specialist users to manifest complex digital infrastructure through high-level natural language intent without compromising on industrial security standards or operational rigor.
Keywords: large language models, Code Generation, WebContainers, cloud deployment, automated software engineering, React
[This article belongs to Journal of Software Engineering Tools & Technology Trends ]
References
- Sommerville I. Software engineering. 10th ed. Boston: Pearson Education Limited; 2016.
- Li Y, Han Z, Zhang Q, Li Z, Tan H. Automating cloud deployment for deep learning inference of real-time online services. IEEE INFOCOM 2020 – IEEE Conference on Computer Communications, Toronto, ON, Canada. 2020. p. 1668-77. doi:10.1109/INFOCOM41043.2020.9155267.
- Jiang J, Wang F, Shen J, Kim S, Kim S. A survey on large language models for code generation. ACM Trans Softw Eng Methodol. 2026;35:1-72. doi:10.1145/3747588.
- Ge Y, Mei L, Duan Z, Li T, Zheng Y, Wang Y, et al. A survey of vibe coding with large language models [Preprint]. 2025 Oct 14. arXiv:2510.12399.
- Brown T, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P, et al. Language models are few-shot learners. Adv Neural Inf Process Syst. 2020;33:1877-1901.
- Zhao J, Sun Y, Huang C, Liu C, Guan Y, Zeng Y, et al. Towards secure code generation with LLMs: a study on common weakness enumeration. IEEE Trans Softw Eng. 2025;51:3507-23. doi:10.1109/TSE.2025.3619281.
- Wu W, Hu H, Fan Z, Qiao Y, Huang Y, Li Y, et al. An empirical study of code clones from commercial AI code generators. Proc ACM Softw Eng. 2025;2:2874-96. doi:10.1145/3729397.
- Matthew UO, Asuni O, Fatai LO. Green software engineering development paradigm: an approach to a sustainable renewable energy future. In: Sharma AK, Chanderwal N, Prajapati A, Singh P, Kansal M, editors. Advancing Software Engineering Through AI, Federated Learning, and Large Language Models. Hershey (PA): IGI Global; 2024. p. 281-94. doi:10.4018/979-8-3693-3502-4.ch018.
- Šimunović I, Rezić K, Franić N, Boduljak G, Batinić M, Jukić I, et al. Can general purpose large language models assist pediatricians in predicting infants with serious bacterial infection? BMC Med Inform Decis Mak. 2025;25:423. doi:10.1186/s12911-025-03258-3. PMID:41239388.
- Simons E. (2021). Introducing WebContainers: Run Node.js natively in your browser. [online] StackBlitz. Available from: https://blog.stackblitz.com/posts/introducing-webcontainers/
- Narkhede N, Shapira G, Palino T. Kafka: the Definitive Guide: Real-Time Data and Stream Processing at Scale. Sebastopol (CA): O’Reilly Media; 2017.
- Raza M, Gulwani S. Automated data extraction using predictive program synthesis. Proc AAAI Conf Artif Intell. 2017;31. doi:10.1609/aaai.v31i1.10668.
Journal of Software Engineering Tools & Technology Trends
| Volume | 13 | |
| Issue | 02 | |
| Received | 08/03/2026 | |
| Accepted | 04/04/2026 | |
| Published | 15/05/2026 | |
| Publication Time | 68 Days |