Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM69209
Volume:
10 |
11 |
Nov 2026
Pages:
1-7
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Artificial Intelligence and Data Science
All India Shri Shivaji Memorial Society's Institute of Information Technology
Artificial Intelligence and Data Science
All India Shri Shivaji Memorial Society's Institute of Information Technology
Artificial Intelligence and Data Science
All India Shri Shivaji Memorial Society's Institute of Information Technology
Artificial Intelligence and Data Science
All India Shri Shivaji Memorial Society's Institute of Information Technology
Artificial Intelligence and Data Science
All India Shri Shivaji Memorial Society's Institute of Information Technology
Modern software repositories couple architecture, database, backend, frontend, API, test, CI/CD and infrastructure artefacts so tightly that no single developer or static tool can keep a complete and current understanding of them. Existing AI coding assistants and single-agent software-engineering (SWE) frameworks perform well within one file or one task, but they keep no persistent model of the repository, so every session starts from a partial and quickly outdated view. This paper presents the Agentic Digital Twin for Software Engineering (ADT-SE), a framework that clones a target GitHub repository and builds a continuously synchronised, queryable digital twin from it using a multi-language Abstract Syntax Tree (AST) parser, a NetworkX code property graph, and a ChromaDB semantic index. A seven-agent Software Development Life Cycle (SDLC) orchestrator implemented in LangGraph — comprising a Product Manager, System Architect, Security Engineer, Full-Stack Developer, QA Tester, Staff Reviewer, and DevOps Engineer — reasons over this twin through a shared typed state object, while a context builder grounds every prompt in graph-neighbourhood and retrieved code context. A bounded self-healing loop returns failing or rejected patches to the Developer agent; accepted changes are committed through a Git workspace manager, proposed on GitHub as pull requests, and re-indexed into the twin, closing the loop between the codebase and its model. A FastAPI gateway streams agent trajectories over server-sent events to a React + Vite dashboard that provides graph exploration, semantic search, and an AST-aware diff and merge studio for human review. This paper describes the architecture, the agent workflow and state model, staged-autonomy governance design, and an evaluation protocol based on SWE-bench-style tasks, seeded vulnerability injection and DevOps Research and Assessment (DORA) metrics against single-agent and manual baselines.
KEYWORDS
This paper presented ADT-SE, a framework that maintains a live digital twin of a software repository and lets a seven-agent LangGraph orchestrator understand, test, review, secure and evolve that repository under human supervision. The combination of a graph-and-vector twin, a shared-state agent society, a bounded self-healing loop, and pull-request delivery with re-indexing closes the loop between code and model — which existing AI coding assistants, single-agent frameworks and passive twins each fail to do. Future work includes: a dedicated Documentation agent and a UX/Accessibility agent for broader lifecycle coverage; twins that span several repositories and microservices for enterprise-scale systems; persistent twin storage using PostgreSQL with a vector extension; reinforcement learning from accepted and rejected decisions to improve agent policies over time; formal verification hooks and stronger sandboxing for fully autonomous production deployment; and packaging ADT-SE as a plug-and-play DevOps add-on compatible with GitHub, GitLab and Bitbucket workflows.
[1] R. Kimmel, J. Michael, A. Wortmann, and J. Zhang, "Digital Twins for Software Engineering Processes," in Proc. 2025 IEEE/ACM 47th Int. Conf. on Software Engineering: New Ideas and Emerging Results (ICSE-NIER), Ottawa, Canada, 2025, doi: 10.1109/ICSE-NIER66352.2025.00009. [2] A. Nechesov, I. Dorokhov, and J. Ruponen, "Virtual Cities: From Digital Twins to Autonomous AI Societies," IEEE Access, vol. 13, pp. 13866–13903, 2025, doi: 10.1109/ACCESS.2025.3531222. [3] M. Baqar, S. Naqvi, and R. Khanda, "AI-Augmented CI/CD Pipelines: From Code Commit to Production with Autonomous Decisions," in Proc. 2025 3rd Int. Conf. on Foundation and Large Language Models (FLLM), 2025, doi: 10.1109/FLLM67465.2025.11391007. [4] V. Tawosi, K. Ramani, S. Alamir, and X. Liu, "ALMAS: an Autonomous LLM-based Multi-Agent Software Engineering Framework," in Proc. 2025 40th IEEE/ACM Int. Conf. on Automated Software Engineering Workshops (ASEW), 2025, pp. 287–290. [5] V. Gogineni, "LLM-Powered Multi-Agent Systems: A Technical Framework for Collaborative Intelligence Through Optimized Knowledge Retrieval and Communication," in Proc. 2025 6th Int. Conf. on Artificial Intelligence, Robotics and Control (AIRC), 2025, pp. 452–456, doi: 10.1109/AIRC64931.2025.11077480. [6] R. J. Tong, H. Li, S. Raghavan, et al., "IEEE AI Standards for Agentic Systems," in Proc. 2025 IEEE Conf. on Artificial Intelligence (CAI), 2025, pp. 1603–1609, doi: 10.1109/CAI64502.2025.00269. [7] W. Liu, A. Yu, D. Zan, B. Shen, W. Zhang, H. Zhao, Z. Jin, and Q. Wang, "GraphCoder: Enhancing Repository-Level Code Completion via Coarse-to-Fine Retrieval Based on Code Context Graph," in Proc. 2024 39th IEEE/ACM Int. Conf. on Automated Software Engineering (ASE), 2024, pp. 570–581, doi: 10.1145/3691620.3695054. [8] J. Yang, C. Jimenez, A. Wettig, K. Lieret, S. Yao, K. Narasimhan, and O. Press, "SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering," in Proc. 38th Conf. on Neural Information Processing Systems (NeurIPS), 2024. [9] C. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, "SWE-bench: Can Language Models Resolve Real-World GitHub Issues?" in Proc. 12th Int. Conf. on Learning Representations (ICLR), 2024. [10] P. Lewis, E. Perez, A. Piktus, et al., "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks," in Proc. 34th Conf. on Neural Information Processing Systems (NeurIPS), 2020.
Publication Date
11/19/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
0
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...